The inclusion of clinical information in diagnostic testing may influence the interpretation of the clinical findings. Historical and clinical findings may focus the reader's attention to the relevant details, thereby improving the accuracy of the interpretation. However, such information may cause the reader to have preconceived notions about the results, biasing the overall interpretation.

The impact of clinical information on the interpretation of radiographic studies remains an issue of debate. Previous studies have found that clinical information improves the accuracy of radiographic interpretation for a broad range of diagnoses,[1, 2, 3, 4] whereas others do not show improvement.[5, 6, 7] Additionally, clinical information may serve as a distraction that leads to more false‐positive interpretations.[8] For this reason, many radiologists prefer to review radiographs without knowledge of the clinical scenario prompting the study to avoid focusing on the expected findings and potentially missing other important abnormalities.[9]

The chest radiograph (CXR) is the most commonly used diagnostic imaging modality. Nevertheless, poor agreement exists among radiologists in the interpretation of chest radiographs for the diagnosis of pneumonia in both adults and children.[10, 11, 12, 13, 14, 15] Recent studies have found a high degree of agreement among pediatric radiologists with implementation of the World Health Organization (WHO) criteria for standardized CXR interpretation for diagnosis of bacterial pneumonia in children.[16, 17, 18] In these studies, participants were blinded to the clinical presentation. Data investigating the impact of clinical history on CXR interpretation in the pediatric population are limited.[19]

We conducted this prospective case‐based study to evaluate the impact of clinical information on the reliability of radiographic diagnosis of pneumonia among children presenting to a pediatric emergency department (ED) with clinical suspicion of pneumonia.

METHODS

RESULTS

Inter‐rater Reliability

The kappa coefficients of inter‐rater reliability between the radiologists across the 6 clinical measures of interest with and without access to clinical history are plotted in Figure 1. Reliability improved from fair (k = 0.32, 95% CI: 0.24 to 0.42) to moderate (k = 0.53, 95% CI: 0.43 to 0.64) for identification of air bronchograms with the addition of clinical history. Although there was an increase in kappa values for identification of any infiltrate, alveolar infiltrate, interstitial infiltrate, and pleural effusion, and a decrease in the kappa value for identification of hilar adenopathy with the addition of clinical information, there was substantial overlap of the 95% CIs, suggesting that inclusion of clinical history did not result in a statistically significant change in the reliability of these findings.

Figure 1

Intra‐rater Concordance

The inter‐rater concordance of the radiologists for the identification of alveolar infiltrate during the interpretation of the 100 chest radiographs with and without access to clinical history is shown in Figure 2. The availability of clinical information impacted physicians differently in the evaluation of alveolar infiltrates. Raters 1, 4, and 6 appeared more likely to identify an alveolar infiltrate with access to the clinical information, whereas raters 3 and 5 appeared less likely to identify an alveolar infiltrate. Of the 100 films that were interpreted with and without clinical information, the mean number of discordant interpretations per rater was 10, with values ranging from 6 to 19 for the individual raters. Radiographs in which more than 3 raters changed their interpretation regarding the presence of an alveolar infiltrate are shown in Figure 3. For Figure 3D, 4 radiologists changed their interpretation from no alveolar infiltrate to alveolar infiltrate, and 1 radiologist changed from alveolar infiltrate to no alveolar infiltrate with the addition of clinical history.

Figure 2

Figure 3

The inclusion of clinical information in diagnostic testing may influence the interpretation of the clinical findings. Historical and clinical findings may focus the reader's attention to the relevant details, thereby improving the accuracy of the interpretation. However, such information may cause the reader to have preconceived notions about the results, biasing the overall interpretation.

The impact of clinical information on the interpretation of radiographic studies remains an issue of debate. Previous studies have found that clinical information improves the accuracy of radiographic interpretation for a broad range of diagnoses,[1, 2, 3, 4] whereas others do not show improvement.[5, 6, 7] Additionally, clinical information may serve as a distraction that leads to more false‐positive interpretations.[8] For this reason, many radiologists prefer to review radiographs without knowledge of the clinical scenario prompting the study to avoid focusing on the expected findings and potentially missing other important abnormalities.[9]

The chest radiograph (CXR) is the most commonly used diagnostic imaging modality. Nevertheless, poor agreement exists among radiologists in the interpretation of chest radiographs for the diagnosis of pneumonia in both adults and children.[10, 11, 12, 13, 14, 15] Recent studies have found a high degree of agreement among pediatric radiologists with implementation of the World Health Organization (WHO) criteria for standardized CXR interpretation for diagnosis of bacterial pneumonia in children.[16, 17, 18] In these studies, participants were blinded to the clinical presentation. Data investigating the impact of clinical history on CXR interpretation in the pediatric population are limited.[19]

We conducted this prospective case‐based study to evaluate the impact of clinical information on the reliability of radiographic diagnosis of pneumonia among children presenting to a pediatric emergency department (ED) with clinical suspicion of pneumonia.

METHODS

RESULTS

Inter‐rater Reliability

The kappa coefficients of inter‐rater reliability between the radiologists across the 6 clinical measures of interest with and without access to clinical history are plotted in Figure 1. Reliability improved from fair (k = 0.32, 95% CI: 0.24 to 0.42) to moderate (k = 0.53, 95% CI: 0.43 to 0.64) for identification of air bronchograms with the addition of clinical history. Although there was an increase in kappa values for identification of any infiltrate, alveolar infiltrate, interstitial infiltrate, and pleural effusion, and a decrease in the kappa value for identification of hilar adenopathy with the addition of clinical information, there was substantial overlap of the 95% CIs, suggesting that inclusion of clinical history did not result in a statistically significant change in the reliability of these findings.

Figure 1

Intra‐rater Concordance

The inter‐rater concordance of the radiologists for the identification of alveolar infiltrate during the interpretation of the 100 chest radiographs with and without access to clinical history is shown in Figure 2. The availability of clinical information impacted physicians differently in the evaluation of alveolar infiltrates. Raters 1, 4, and 6 appeared more likely to identify an alveolar infiltrate with access to the clinical information, whereas raters 3 and 5 appeared less likely to identify an alveolar infiltrate. Of the 100 films that were interpreted with and without clinical information, the mean number of discordant interpretations per rater was 10, with values ranging from 6 to 19 for the individual raters. Radiographs in which more than 3 raters changed their interpretation regarding the presence of an alveolar infiltrate are shown in Figure 3. For Figure 3D, 4 radiologists changed their interpretation from no alveolar infiltrate to alveolar infiltrate, and 1 radiologist changed from alveolar infiltrate to no alveolar infiltrate with the addition of clinical history.

Figure 2

Figure 3

The inclusion of clinical information in diagnostic testing may influence the interpretation of the clinical findings. Historical and clinical findings may focus the reader's attention to the relevant details, thereby improving the accuracy of the interpretation. However, such information may cause the reader to have preconceived notions about the results, biasing the overall interpretation.

The impact of clinical information on the interpretation of radiographic studies remains an issue of debate. Previous studies have found that clinical information improves the accuracy of radiographic interpretation for a broad range of diagnoses,[1, 2, 3, 4] whereas others do not show improvement.[5, 6, 7] Additionally, clinical information may serve as a distraction that leads to more false‐positive interpretations.[8] For this reason, many radiologists prefer to review radiographs without knowledge of the clinical scenario prompting the study to avoid focusing on the expected findings and potentially missing other important abnormalities.[9]

The chest radiograph (CXR) is the most commonly used diagnostic imaging modality. Nevertheless, poor agreement exists among radiologists in the interpretation of chest radiographs for the diagnosis of pneumonia in both adults and children.[10, 11, 12, 13, 14, 15] Recent studies have found a high degree of agreement among pediatric radiologists with implementation of the World Health Organization (WHO) criteria for standardized CXR interpretation for diagnosis of bacterial pneumonia in children.[16, 17, 18] In these studies, participants were blinded to the clinical presentation. Data investigating the impact of clinical history on CXR interpretation in the pediatric population are limited.[19]

We conducted this prospective case‐based study to evaluate the impact of clinical information on the reliability of radiographic diagnosis of pneumonia among children presenting to a pediatric emergency department (ED) with clinical suspicion of pneumonia.

METHODS

RESULTS

Inter‐rater Reliability

The kappa coefficients of inter‐rater reliability between the radiologists across the 6 clinical measures of interest with and without access to clinical history are plotted in Figure 1. Reliability improved from fair (k = 0.32, 95% CI: 0.24 to 0.42) to moderate (k = 0.53, 95% CI: 0.43 to 0.64) for identification of air bronchograms with the addition of clinical history. Although there was an increase in kappa values for identification of any infiltrate, alveolar infiltrate, interstitial infiltrate, and pleural effusion, and a decrease in the kappa value for identification of hilar adenopathy with the addition of clinical information, there was substantial overlap of the 95% CIs, suggesting that inclusion of clinical history did not result in a statistically significant change in the reliability of these findings.

Figure 1

Intra‐rater Concordance

The inter‐rater concordance of the radiologists for the identification of alveolar infiltrate during the interpretation of the 100 chest radiographs with and without access to clinical history is shown in Figure 2. The availability of clinical information impacted physicians differently in the evaluation of alveolar infiltrates. Raters 1, 4, and 6 appeared more likely to identify an alveolar infiltrate with access to the clinical information, whereas raters 3 and 5 appeared less likely to identify an alveolar infiltrate. Of the 100 films that were interpreted with and without clinical information, the mean number of discordant interpretations per rater was 10, with values ranging from 6 to 19 for the individual raters. Radiographs in which more than 3 raters changed their interpretation regarding the presence of an alveolar infiltrate are shown in Figure 3. For Figure 3D, 4 radiologists changed their interpretation from no alveolar infiltrate to alveolar infiltrate, and 1 radiologist changed from alveolar infiltrate to no alveolar infiltrate with the addition of clinical history.

Figure 2

Figure 3

Most patients prefer to die at home, pain free and without the use of life‐sustaining treatments[1, 2] yet the majority of patients with serious illness die in the hospital.[3, 4, 5] When hospitalized patient die, the care is frequently focused on life‐sustaining treatments[6]; pain, dyspnea, and agitation levels are higher when compared with patients who die in non‐hospital settings.[7, 8, 9] End‐of‐life discussions and their products (ie, advanced directives) can clarify treatment options with patients and family,[10] and help ensure that patients receive care consistent with their beliefs.[11, 12, 13] End‐of‐life discussions are associated with a decrease use of life‐sustaining treatments, improved quality of life, and reduced costs of care.[14, 15] For the majority of patients dying of cancer, the first end‐of‐life discussion takes place in the hospital setting.[16]

Conducting end‐of‐life conversations in the hospital setting can be challenging. Patients are acutely ill and nearly 40% are incapable of making their own medical decisions.[17] In order to participate in an end‐of‐life discussion, a physician must determine that a patient meets the 4 criteria of decisional capacity as outlined by Appelbaum and Grisso:[18] Does the patient (1) communicate a clear and consistent choice; (2) understand the relevant information surrounding that decision; (3) appreciate the consequences of that decision; and (4) communicate reasoning for that decision?[19] In practice, however, clinicians inaccurately assign capacity up to 25% of the time.[17] When a physician determines, accurately or inaccurately, that the patient does not meet this standard for decisional capacity, discussions must be held instead with a surrogate decision‐maker. Surrogate decision‐making can make communication with the physician more difficult,[20] delay important medical decisions,[21] and be stressful on the decision‐maker.[22] To our knowledge, no studies have examined patient and surrogate participation in end‐of‐life discussions at the time of terminal hospitalization and its association with end‐of‐life treatments received.

Our goals were to examine physician assessment of decisional capacity and the prevalence of end‐of‐life discussions during the terminal hospitalization of patients with advanced cancer. Our research questions were: (1) What proportion of patients were assessed to have decisional capacity by the clinical team at the time of their terminal hospital admission? (2) What proportion of these patients had a documented discussion about end‐of‐life care with the clinical team, and for what proportion was the conversation held instead by the patient's surrogate decision‐maker because the patient was considered to have lost decisional capacity? (3) Was patient participation in a discussion about end‐of‐life care associated with life‐sustaining and palliative treatments received?

METHODS

RESULTS

DISCUSSION

In this retrospective study of 145 terminal cancer patients who died in the hospital, we found that half of the patients had documentation suggesting that they lost decisional capacity during hospitalization and did not participate in end‐of‐life discussion with their healthcare providers. Among cancer patients in our study, 19% were determined not to have decisional capacity on admission, and another 32% were determined to lose decisional capacity during the hospital course. When patients did participate in documented discussions about end‐of‐life care, they were less likely to receive intensive life‐sustaining treatments, less likely to be admitted or transferred to the ICU, and more likely to avoid prolonged hospitalization. The finding that the majority of cancer patients are assessed to have intact decisional capacity upon admission to their terminal hospitalization, but less than half of them participate in their own end‐of‐life conversation, suggests that there is an important lost opportunity to provide quality advanced care planning to hospitalized patients.

Advanced care planning is the act of defining a competent patient's wishes regarding their future healthcare in the event of loss of decisional capacity. For many of the patients who were determined to lose decisional capacity during their hospitalization in our study, a surrogate decision‐maker was involved in a subsequent end‐of‐life discussion. This represents a missed opportunity in 2 ways. First, because surrogates may incorrectly predict patients' end‐of‐life treatment preferences in approximately one‐third of the cases,[25] patients may receive care that is inconsistent with their beliefs. Second, reliance on surrogate decision‐making may result in a greater burden on family members. Surrogate decision‐making places a large burden on surrogates, and can lead to emotional and psychological stress that can last well beyond the death of the loved one.[16, 26]

Our results reinforce the growing body of evidence suggesting that communication with the dying patient, both before and during the terminal hospitalization, promotes end‐of‐life care that involves less invasive life‐prolonging treatments,[10, 11, 12, 13, 14, 15] which is a consistently stated goal of most patients at the end of life.[1, 2] Our findings are consistent with a recent randomized trial of hospitalized patients over the age of 80 which showed that advance care planning discussions were associated with decreased use of intensive life‐sustaining treatments, increased patient and family satisfaction with care, as well as decreased psychological symptoms among family members.[27] In addition, studies examining patients with advanced cancer have shown that discussions about end‐of‐life care in the outpatient setting resulted in less intensive life‐sustaining treatments.[14, 15] Nearly 20% of our cohort was determined by the clinical team to lack decisional capacity at the time of their terminal hospitalization, highlighting the importance of end‐of‐life discussion prior to hospitalization. Our results also extend these recent findings to the inpatient setting. Seventy‐nine percent of patients in our study were determined to have decisional capacity at the time of their terminal hospitalization, and end‐of‐life discussion conducted with these patients was associated with decreased use of life‐sustaining treatments.

Seriously ill patients, particularly in hospital settings, have high symptom burden and subsequent poor quality of life.[6] These patients and families often report inadequate pain and symptom relief.[8, 28, 29] It is therefore reassuring that we found no difference in the proportion of patients who received comfort care and palliative morphine infusions according to whether they or a surrogate participated in the end‐of‐life discussions. In fact, the majority of patients were made DNR and received some form of comfort measure prior to death, although these comfort measures were frequently initiated only hours prior to death.

While our findings are consistent with research that has demonstrated that end‐of‐life discussions with patients are associated with a decrease in life‐prolonging treatments, it is important to note that our observational study cannot establish a causal relationship between the end‐of‐life discussion and the subsequent use of life‐sustaining treatments. It is possible that patients who have discussions about end‐of‐life care inherently prefer to have less intensive life‐sustaining treatment at the end of life. Physicians may be more apt to engage in these types of discussions with patients who express interest in limited intervention. Interestingly, patients with a DNR order at the time of admission were more likely to participate in end‐of‐life care with their provider during their terminal hospitalization, which supports this alternate explanation. However, when we excluded all patients with a DNR on admission, our findings persisted. Regardless of the explanation for the association between end‐of‐life discussions and life‐sustaining treatments, our study identifies a cohort of hospitalized patients who could benefit from improved end‐of‐life communication, and a clinical setting where opportunities remain to improve the quality of advanced care planning.

Discussions about end‐of‐life care with patients result in earlier transition to care focused on palliation.[14] Although we examined the timing to initiation of comfort care between those who participated and those who did not participate in end‐of‐life discussions, we were not able to demonstrate a statistically significant difference. However, our cohort may not be suitable for an examination of this type of intervention, as early discussions about end‐of‐life care may have lead to early referral to hospice, and therefore death outside the hospital setting, making the patients ineligible for our study.

There are several additional limitations to our study. First, our study is subject to the potential biases inherent in retrospective chart review. For example, since we identified patients who died in the hospital, our study does not generalize to patients with advanced cancer who survive the hospitalization or who were discharged to die at home.[30] Furthermore, our use of the medical record to assess documentation of communication about end‐of‐life care and decisional capacity relies on the accuracy of clinician documentation. There may have been communication about end‐of‐life care that was not documented in the medical record, resulting in underestimation of the effects of end‐of‐life discussions. In clinical practice, the assessment of decisional capacity can be challenging.[19, 31, 32] Although clinicians are often accurate in identifying patients with capacity, in nearly one‐quarter of all assessments, they mistakenly assign capacity to patients who lack decision‐making capacity.[17] In our study, we examine clinician assessment of the patient's decisional capacity, but cannot assess either the accuracy of their clinical assessment or the accuracy of their documentation of this assessment. A second limitation is that the chart abstraction was conducted by a single reviewer without inter‐rater assessment, although the abstraction tool was modified from a well‐established tool.[24] A third limitation is that all of the patients were from a single academic center and the results may not generalize to other regions. Fourth, we did not assess for the potential effect of individual clinicians. Since most patients were cared for by multiple physicians, our study is not capable of assessing a clinician‐level effect. Finally, our study was not able to address whether the care received by patients was in accord with their informed preferences.

We have shown that opportunities exist for advanced care planning at the time of the terminal hospitalization of patients with advanced cancer. The majority of these patients have documentation of decisional capacity at the time of admission, suggesting that opportunities exist to conduct end‐of‐life discussion while the patient retains decisional capacity. Furthermore, we found that patients who participate in these discussions about end‐of‐life care with their clinicians have an associated decrease in the use of life‐sustaining treatments, which is consistent with prior studies.[11, 13, 14, 15, 27] Improving communication about end‐of‐life care for patients hospitalized with advanced cancer may represent an important opportunity to improve the concordance between patients' wishes for care at the end‐of‐life and the care that these patients actually receive. Such communication may also decrease the burden on family members who are frequently asked to play the role of surrogate decision‐maker without an opportunity to discuss these issues with the patient.

Most patients prefer to die at home, pain free and without the use of life‐sustaining treatments[1, 2] yet the majority of patients with serious illness die in the hospital.[3, 4, 5] When hospitalized patient die, the care is frequently focused on life‐sustaining treatments[6]; pain, dyspnea, and agitation levels are higher when compared with patients who die in non‐hospital settings.[7, 8, 9] End‐of‐life discussions and their products (ie, advanced directives) can clarify treatment options with patients and family,[10] and help ensure that patients receive care consistent with their beliefs.[11, 12, 13] End‐of‐life discussions are associated with a decrease use of life‐sustaining treatments, improved quality of life, and reduced costs of care.[14, 15] For the majority of patients dying of cancer, the first end‐of‐life discussion takes place in the hospital setting.[16]

Conducting end‐of‐life conversations in the hospital setting can be challenging. Patients are acutely ill and nearly 40% are incapable of making their own medical decisions.[17] In order to participate in an end‐of‐life discussion, a physician must determine that a patient meets the 4 criteria of decisional capacity as outlined by Appelbaum and Grisso:[18] Does the patient (1) communicate a clear and consistent choice; (2) understand the relevant information surrounding that decision; (3) appreciate the consequences of that decision; and (4) communicate reasoning for that decision?[19] In practice, however, clinicians inaccurately assign capacity up to 25% of the time.[17] When a physician determines, accurately or inaccurately, that the patient does not meet this standard for decisional capacity, discussions must be held instead with a surrogate decision‐maker. Surrogate decision‐making can make communication with the physician more difficult,[20] delay important medical decisions,[21] and be stressful on the decision‐maker.[22] To our knowledge, no studies have examined patient and surrogate participation in end‐of‐life discussions at the time of terminal hospitalization and its association with end‐of‐life treatments received.

Our goals were to examine physician assessment of decisional capacity and the prevalence of end‐of‐life discussions during the terminal hospitalization of patients with advanced cancer. Our research questions were: (1) What proportion of patients were assessed to have decisional capacity by the clinical team at the time of their terminal hospital admission? (2) What proportion of these patients had a documented discussion about end‐of‐life care with the clinical team, and for what proportion was the conversation held instead by the patient's surrogate decision‐maker because the patient was considered to have lost decisional capacity? (3) Was patient participation in a discussion about end‐of‐life care associated with life‐sustaining and palliative treatments received?

METHODS

RESULTS

DISCUSSION

In this retrospective study of 145 terminal cancer patients who died in the hospital, we found that half of the patients had documentation suggesting that they lost decisional capacity during hospitalization and did not participate in end‐of‐life discussion with their healthcare providers. Among cancer patients in our study, 19% were determined not to have decisional capacity on admission, and another 32% were determined to lose decisional capacity during the hospital course. When patients did participate in documented discussions about end‐of‐life care, they were less likely to receive intensive life‐sustaining treatments, less likely to be admitted or transferred to the ICU, and more likely to avoid prolonged hospitalization. The finding that the majority of cancer patients are assessed to have intact decisional capacity upon admission to their terminal hospitalization, but less than half of them participate in their own end‐of‐life conversation, suggests that there is an important lost opportunity to provide quality advanced care planning to hospitalized patients.

Advanced care planning is the act of defining a competent patient's wishes regarding their future healthcare in the event of loss of decisional capacity. For many of the patients who were determined to lose decisional capacity during their hospitalization in our study, a surrogate decision‐maker was involved in a subsequent end‐of‐life discussion. This represents a missed opportunity in 2 ways. First, because surrogates may incorrectly predict patients' end‐of‐life treatment preferences in approximately one‐third of the cases,[25] patients may receive care that is inconsistent with their beliefs. Second, reliance on surrogate decision‐making may result in a greater burden on family members. Surrogate decision‐making places a large burden on surrogates, and can lead to emotional and psychological stress that can last well beyond the death of the loved one.[16, 26]

Our results reinforce the growing body of evidence suggesting that communication with the dying patient, both before and during the terminal hospitalization, promotes end‐of‐life care that involves less invasive life‐prolonging treatments,[10, 11, 12, 13, 14, 15] which is a consistently stated goal of most patients at the end of life.[1, 2] Our findings are consistent with a recent randomized trial of hospitalized patients over the age of 80 which showed that advance care planning discussions were associated with decreased use of intensive life‐sustaining treatments, increased patient and family satisfaction with care, as well as decreased psychological symptoms among family members.[27] In addition, studies examining patients with advanced cancer have shown that discussions about end‐of‐life care in the outpatient setting resulted in less intensive life‐sustaining treatments.[14, 15] Nearly 20% of our cohort was determined by the clinical team to lack decisional capacity at the time of their terminal hospitalization, highlighting the importance of end‐of‐life discussion prior to hospitalization. Our results also extend these recent findings to the inpatient setting. Seventy‐nine percent of patients in our study were determined to have decisional capacity at the time of their terminal hospitalization, and end‐of‐life discussion conducted with these patients was associated with decreased use of life‐sustaining treatments.

Seriously ill patients, particularly in hospital settings, have high symptom burden and subsequent poor quality of life.[6] These patients and families often report inadequate pain and symptom relief.[8, 28, 29] It is therefore reassuring that we found no difference in the proportion of patients who received comfort care and palliative morphine infusions according to whether they or a surrogate participated in the end‐of‐life discussions. In fact, the majority of patients were made DNR and received some form of comfort measure prior to death, although these comfort measures were frequently initiated only hours prior to death.

While our findings are consistent with research that has demonstrated that end‐of‐life discussions with patients are associated with a decrease in life‐prolonging treatments, it is important to note that our observational study cannot establish a causal relationship between the end‐of‐life discussion and the subsequent use of life‐sustaining treatments. It is possible that patients who have discussions about end‐of‐life care inherently prefer to have less intensive life‐sustaining treatment at the end of life. Physicians may be more apt to engage in these types of discussions with patients who express interest in limited intervention. Interestingly, patients with a DNR order at the time of admission were more likely to participate in end‐of‐life care with their provider during their terminal hospitalization, which supports this alternate explanation. However, when we excluded all patients with a DNR on admission, our findings persisted. Regardless of the explanation for the association between end‐of‐life discussions and life‐sustaining treatments, our study identifies a cohort of hospitalized patients who could benefit from improved end‐of‐life communication, and a clinical setting where opportunities remain to improve the quality of advanced care planning.

Discussions about end‐of‐life care with patients result in earlier transition to care focused on palliation.[14] Although we examined the timing to initiation of comfort care between those who participated and those who did not participate in end‐of‐life discussions, we were not able to demonstrate a statistically significant difference. However, our cohort may not be suitable for an examination of this type of intervention, as early discussions about end‐of‐life care may have lead to early referral to hospice, and therefore death outside the hospital setting, making the patients ineligible for our study.

There are several additional limitations to our study. First, our study is subject to the potential biases inherent in retrospective chart review. For example, since we identified patients who died in the hospital, our study does not generalize to patients with advanced cancer who survive the hospitalization or who were discharged to die at home.[30] Furthermore, our use of the medical record to assess documentation of communication about end‐of‐life care and decisional capacity relies on the accuracy of clinician documentation. There may have been communication about end‐of‐life care that was not documented in the medical record, resulting in underestimation of the effects of end‐of‐life discussions. In clinical practice, the assessment of decisional capacity can be challenging.[19, 31, 32] Although clinicians are often accurate in identifying patients with capacity, in nearly one‐quarter of all assessments, they mistakenly assign capacity to patients who lack decision‐making capacity.[17] In our study, we examine clinician assessment of the patient's decisional capacity, but cannot assess either the accuracy of their clinical assessment or the accuracy of their documentation of this assessment. A second limitation is that the chart abstraction was conducted by a single reviewer without inter‐rater assessment, although the abstraction tool was modified from a well‐established tool.[24] A third limitation is that all of the patients were from a single academic center and the results may not generalize to other regions. Fourth, we did not assess for the potential effect of individual clinicians. Since most patients were cared for by multiple physicians, our study is not capable of assessing a clinician‐level effect. Finally, our study was not able to address whether the care received by patients was in accord with their informed preferences.

We have shown that opportunities exist for advanced care planning at the time of the terminal hospitalization of patients with advanced cancer. The majority of these patients have documentation of decisional capacity at the time of admission, suggesting that opportunities exist to conduct end‐of‐life discussion while the patient retains decisional capacity. Furthermore, we found that patients who participate in these discussions about end‐of‐life care with their clinicians have an associated decrease in the use of life‐sustaining treatments, which is consistent with prior studies.[11, 13, 14, 15, 27] Improving communication about end‐of‐life care for patients hospitalized with advanced cancer may represent an important opportunity to improve the concordance between patients' wishes for care at the end‐of‐life and the care that these patients actually receive. Such communication may also decrease the burden on family members who are frequently asked to play the role of surrogate decision‐maker without an opportunity to discuss these issues with the patient.

Most patients prefer to die at home, pain free and without the use of life‐sustaining treatments[1, 2] yet the majority of patients with serious illness die in the hospital.[3, 4, 5] When hospitalized patient die, the care is frequently focused on life‐sustaining treatments[6]; pain, dyspnea, and agitation levels are higher when compared with patients who die in non‐hospital settings.[7, 8, 9] End‐of‐life discussions and their products (ie, advanced directives) can clarify treatment options with patients and family,[10] and help ensure that patients receive care consistent with their beliefs.[11, 12, 13] End‐of‐life discussions are associated with a decrease use of life‐sustaining treatments, improved quality of life, and reduced costs of care.[14, 15] For the majority of patients dying of cancer, the first end‐of‐life discussion takes place in the hospital setting.[16]

Conducting end‐of‐life conversations in the hospital setting can be challenging. Patients are acutely ill and nearly 40% are incapable of making their own medical decisions.[17] In order to participate in an end‐of‐life discussion, a physician must determine that a patient meets the 4 criteria of decisional capacity as outlined by Appelbaum and Grisso:[18] Does the patient (1) communicate a clear and consistent choice; (2) understand the relevant information surrounding that decision; (3) appreciate the consequences of that decision; and (4) communicate reasoning for that decision?[19] In practice, however, clinicians inaccurately assign capacity up to 25% of the time.[17] When a physician determines, accurately or inaccurately, that the patient does not meet this standard for decisional capacity, discussions must be held instead with a surrogate decision‐maker. Surrogate decision‐making can make communication with the physician more difficult,[20] delay important medical decisions,[21] and be stressful on the decision‐maker.[22] To our knowledge, no studies have examined patient and surrogate participation in end‐of‐life discussions at the time of terminal hospitalization and its association with end‐of‐life treatments received.

Our goals were to examine physician assessment of decisional capacity and the prevalence of end‐of‐life discussions during the terminal hospitalization of patients with advanced cancer. Our research questions were: (1) What proportion of patients were assessed to have decisional capacity by the clinical team at the time of their terminal hospital admission? (2) What proportion of these patients had a documented discussion about end‐of‐life care with the clinical team, and for what proportion was the conversation held instead by the patient's surrogate decision‐maker because the patient was considered to have lost decisional capacity? (3) Was patient participation in a discussion about end‐of‐life care associated with life‐sustaining and palliative treatments received?

METHODS

RESULTS

DISCUSSION

In this retrospective study of 145 terminal cancer patients who died in the hospital, we found that half of the patients had documentation suggesting that they lost decisional capacity during hospitalization and did not participate in end‐of‐life discussion with their healthcare providers. Among cancer patients in our study, 19% were determined not to have decisional capacity on admission, and another 32% were determined to lose decisional capacity during the hospital course. When patients did participate in documented discussions about end‐of‐life care, they were less likely to receive intensive life‐sustaining treatments, less likely to be admitted or transferred to the ICU, and more likely to avoid prolonged hospitalization. The finding that the majority of cancer patients are assessed to have intact decisional capacity upon admission to their terminal hospitalization, but less than half of them participate in their own end‐of‐life conversation, suggests that there is an important lost opportunity to provide quality advanced care planning to hospitalized patients.

Advanced care planning is the act of defining a competent patient's wishes regarding their future healthcare in the event of loss of decisional capacity. For many of the patients who were determined to lose decisional capacity during their hospitalization in our study, a surrogate decision‐maker was involved in a subsequent end‐of‐life discussion. This represents a missed opportunity in 2 ways. First, because surrogates may incorrectly predict patients' end‐of‐life treatment preferences in approximately one‐third of the cases,[25] patients may receive care that is inconsistent with their beliefs. Second, reliance on surrogate decision‐making may result in a greater burden on family members. Surrogate decision‐making places a large burden on surrogates, and can lead to emotional and psychological stress that can last well beyond the death of the loved one.[16, 26]

Our results reinforce the growing body of evidence suggesting that communication with the dying patient, both before and during the terminal hospitalization, promotes end‐of‐life care that involves less invasive life‐prolonging treatments,[10, 11, 12, 13, 14, 15] which is a consistently stated goal of most patients at the end of life.[1, 2] Our findings are consistent with a recent randomized trial of hospitalized patients over the age of 80 which showed that advance care planning discussions were associated with decreased use of intensive life‐sustaining treatments, increased patient and family satisfaction with care, as well as decreased psychological symptoms among family members.[27] In addition, studies examining patients with advanced cancer have shown that discussions about end‐of‐life care in the outpatient setting resulted in less intensive life‐sustaining treatments.[14, 15] Nearly 20% of our cohort was determined by the clinical team to lack decisional capacity at the time of their terminal hospitalization, highlighting the importance of end‐of‐life discussion prior to hospitalization. Our results also extend these recent findings to the inpatient setting. Seventy‐nine percent of patients in our study were determined to have decisional capacity at the time of their terminal hospitalization, and end‐of‐life discussion conducted with these patients was associated with decreased use of life‐sustaining treatments.

Seriously ill patients, particularly in hospital settings, have high symptom burden and subsequent poor quality of life.[6] These patients and families often report inadequate pain and symptom relief.[8, 28, 29] It is therefore reassuring that we found no difference in the proportion of patients who received comfort care and palliative morphine infusions according to whether they or a surrogate participated in the end‐of‐life discussions. In fact, the majority of patients were made DNR and received some form of comfort measure prior to death, although these comfort measures were frequently initiated only hours prior to death.

While our findings are consistent with research that has demonstrated that end‐of‐life discussions with patients are associated with a decrease in life‐prolonging treatments, it is important to note that our observational study cannot establish a causal relationship between the end‐of‐life discussion and the subsequent use of life‐sustaining treatments. It is possible that patients who have discussions about end‐of‐life care inherently prefer to have less intensive life‐sustaining treatment at the end of life. Physicians may be more apt to engage in these types of discussions with patients who express interest in limited intervention. Interestingly, patients with a DNR order at the time of admission were more likely to participate in end‐of‐life care with their provider during their terminal hospitalization, which supports this alternate explanation. However, when we excluded all patients with a DNR on admission, our findings persisted. Regardless of the explanation for the association between end‐of‐life discussions and life‐sustaining treatments, our study identifies a cohort of hospitalized patients who could benefit from improved end‐of‐life communication, and a clinical setting where opportunities remain to improve the quality of advanced care planning.

Discussions about end‐of‐life care with patients result in earlier transition to care focused on palliation.[14] Although we examined the timing to initiation of comfort care between those who participated and those who did not participate in end‐of‐life discussions, we were not able to demonstrate a statistically significant difference. However, our cohort may not be suitable for an examination of this type of intervention, as early discussions about end‐of‐life care may have lead to early referral to hospice, and therefore death outside the hospital setting, making the patients ineligible for our study.

There are several additional limitations to our study. First, our study is subject to the potential biases inherent in retrospective chart review. For example, since we identified patients who died in the hospital, our study does not generalize to patients with advanced cancer who survive the hospitalization or who were discharged to die at home.[30] Furthermore, our use of the medical record to assess documentation of communication about end‐of‐life care and decisional capacity relies on the accuracy of clinician documentation. There may have been communication about end‐of‐life care that was not documented in the medical record, resulting in underestimation of the effects of end‐of‐life discussions. In clinical practice, the assessment of decisional capacity can be challenging.[19, 31, 32] Although clinicians are often accurate in identifying patients with capacity, in nearly one‐quarter of all assessments, they mistakenly assign capacity to patients who lack decision‐making capacity.[17] In our study, we examine clinician assessment of the patient's decisional capacity, but cannot assess either the accuracy of their clinical assessment or the accuracy of their documentation of this assessment. A second limitation is that the chart abstraction was conducted by a single reviewer without inter‐rater assessment, although the abstraction tool was modified from a well‐established tool.[24] A third limitation is that all of the patients were from a single academic center and the results may not generalize to other regions. Fourth, we did not assess for the potential effect of individual clinicians. Since most patients were cared for by multiple physicians, our study is not capable of assessing a clinician‐level effect. Finally, our study was not able to address whether the care received by patients was in accord with their informed preferences.

We have shown that opportunities exist for advanced care planning at the time of the terminal hospitalization of patients with advanced cancer. The majority of these patients have documentation of decisional capacity at the time of admission, suggesting that opportunities exist to conduct end‐of‐life discussion while the patient retains decisional capacity. Furthermore, we found that patients who participate in these discussions about end‐of‐life care with their clinicians have an associated decrease in the use of life‐sustaining treatments, which is consistent with prior studies.[11, 13, 14, 15, 27] Improving communication about end‐of‐life care for patients hospitalized with advanced cancer may represent an important opportunity to improve the concordance between patients' wishes for care at the end‐of‐life and the care that these patients actually receive. Such communication may also decrease the burden on family members who are frequently asked to play the role of surrogate decision‐maker without an opportunity to discuss these issues with the patient.

Person‐organization fit concerns the conditions and consequences of compatibility between people and the organizations for which they work.[1] Studies of other industries have demonstrated that person‐organization fit informs the way individuals join, perform in, and are retained by organizations.[2] Person‐job fit is a closely related subordinate concept that concerns the alignment of workers and their job in as much as workers have needs that their job supplies, or conversely, jobs have requirements that certain workers' abilities can help meet.[3] Explorations of job fit in physicians and their work have recently emerged in a few investigations published in medical journals.[4, 5, 6, 7, 8] Further expanding the understanding of fit between physicians and their employment is important, because the decline of solo practices and recent emphasis on team‐based care have led to a growing number of US physicians working in organizations.[9]

The movement of physicians into employed situations may continue if certain types of Accountable Care Organizations take root.[10] And physicians may be primed to join employer organizations based on current career priorities of individuals in American society. Surveys of medical residents entering the workforce reveal more physicians preferring the security of being employees than starting their own practices.[11] Given these trends, job fit will inform our understanding of how personal and job characteristics facilitate recruitment, performance, satisfaction, and longevity of physician employees.

BACKGROUND

Virtually all hospitalists work in organizationshospitalsand are employees of hospitals, medical schools, physician group practices, or management companies, and therefore invariably function within organizational structures and systems.[7] In spite of their rapid growth in numbers, many employers have faced difficulties recruiting and retaining enough hospitalists to fill their staffing needs. Consequently, the US hospitalist workforce today is characterized by high salaries, work load, and attrition rates.[12]

In this evolving unsaturated market, the attraction‐selection‐attrition framework[13] provides a theoretical construct that predicts that hospitalists and their employers would seek congruence of goals and values early in their relationship through a process of trial and error. This framework assumes that early interactions between workers and organizations serve as opportunities for them to understand if job fit is poor and dissociate or remain affiliated as long as job fit is mutually acceptable. Therefore, job switching on average is expected to increase job fit because workers and organizations gain a better understanding of their own goals and values and choose more wisely the next time.

Other theoretical frameworks, such as the job characteristic model,[14] suggest that over time as workers stay at the same job, they continue to maintain and improve job fit through various workplace‐ or self‐modification strategies. For example, seniority status may have privileges (eg, less undesirable call), or workers may create privileged niches through the acquisition of new skills and abilities over time. Hospitalists' tendency to diversify their work‐related activities by incorporating administrative and teaching responsibilities[15] may thus contribute to improving job fit. Additionally, as a measure of complementarity among people who work together, job fit may be influenced by the quality of relationships among hospitalists and their coworkers through their reorientation to the prevailing organizational climate[16, 17] and increasing socialization.[18] Finally, given that experiential learning is known to contribute to better hospitalist work performance,[19] job fit may affect productivity and clinical outcomes vis‐‐vis quality of work life.

To test the validity of these assumptions in a sample of hospitalists, we critically appraised the following 4 hypotheses:

• Hypothesis 1 (H1): Job attrition and reselection improves job fit among hospitalists entering the job market.
• Hypothesis 2 (H2): Better job fit is achieved through hospitalists engaging a variety of personal skills and abilities.
• Hypothesis 3 (H3): Job fit increases with hospitalists' job duration together with socialization and internalization of organizational values.
• Hypothesis 4 (H4): Job fit is correlated with hospitalists' quality of work life.

METHODS

Analysis was performed on data from the 2009 to 2010 Hospital Medicine Physician Worklife Survey. The sample frame included nonmembers and members of Society of Hospital Medicine (SHM). Details about sampling strategy, data collection, and data quality are available in previous publications.[7, 20] The 118‐item survey instrument, including 9 demographic items and 24 practice and job characteristic items, was administered by mail. Examples of information solicited through these items included respondents' practice model, the number of hospitalist jobs they have held, and the specific kinds of clinical and nonclinical activities they performed as part of their current job.

We used a reliable but broad and generic measure of self‐perceived person‐job fit.[21] The survey items of the 5‐point Likert‐type scale anchored between strongly disagree and strongly agree were: I feel that my work utilizes my full abilities, I feel competent and fully able to handle my job, my job gives me the chance to do the things I feel I do best, I feel that my job and I are well‐matched, I feel I have adequate preparation for the job I now hold. The quality of hospitalists' relationships with physician colleagues, staff, and patients as well as job satisfaction was measured using scales adapted from the Physician Worklife Study.[22] Organizational climate was measured using an adapted scale from the Minimizing Error, Maximizing Outcome study incorporating 3 items from the cohesiveness subscale, 4 items from the organizational trust subscale, and 1 item from the quality emphasis subscale that were most pertinent to hospitalists' relationship with their organizations.[23] Intent to leave practice or reduce work hours was measured using 5 items from the Multi‐Center Hospitalist Survey Project.[24] Frequency of participation in suboptimal patient care was measured by adapting 3 items from the suboptimal reported practice subscale and 2 items from the suboptimal patient care subscale developed by Shanafelt et al.[25] Stress and job burnout were assessed using validated measures.[26, 27] Detailed descriptions of the response rate calculation and imputation of missing item data are available in previous publications.[7, 20]

Mean, variance, range, and skew were used to characterize the responses to the job fit survey scale. A table of respondent characteristics was constructed. A visual representation of job fit by individual hospitalist year in current practice was created, first, by plotting a locally weighted scatterplot smoothing curve to examine the shape of the general relationship, and second, by fitting a similarly contoured functional polynomial curve with 95% confidence intervals (CI) to a plot of the mean and interquartile range of job fit for each year in current practice. Spearman partial correlations were calculated for job fit and each of the 5 items addressing likelihood of leaving practice or reducing workload adjusted for gender to control for the higher proportion of women who plan to work part time. Median (interquartile range) job fit was calculated for categories defined by the number of job changes and compared with the reference category (no job change) using the nonparametric rank sum test for comparing non‐normally distributed data. Multivariate logistic regression models were used to calculate the odds ratio (OR) of participating in each of several clinical and nonclinical hospitalist activities between respondents whose job fit score was optimal (5 on a 5‐point scale) and less than optimal controlling for covariates that influence the likelihood of participating in these activities (years in current practice, practice model, and specialty training). A Spearman correlation matrix was created to assess interscale correlations among organizational parameters (years in current practice, job fit, organizational climate, and relationship with colleagues, staff, and patients). Finally, a separate Spearman correlation matrix was created to assess the interscale correlations among individual worker parameters (job fit, suboptimal patient care, job burnout, stress, and job satisfaction). Statistical significance was defined as P value <0.05, and all analyses were performed on Stata 11.0 (StataCorp, College Station, TX). The Northwestern University institutional review board approved this study.

RESULTS

Respondents included 816 hospitalists belonging to around 700 unique organizations. The adjusted response rate from the stratified sample was 26%. Respondents and nonrespondents were similar with regard to geographic region and model of practice, but respondents were more likely to be members of the SHM than nonrespondents. Panel A of Table 1 shows the demographic characteristics of the respondents. The mean age was 44.3 years, and about one‐third were women. The average hospitalist had about 7 years of experience in the specialty and about 5 years with their current hospitalist job. The majority were trained in internal medicine or one of its subspecialties, whereas pediatricians, family physicians, and physicians with other training made up the remainder.

Job fit was highly skewed toward optimum fit, with a mean of 4.3 on a scale of 1 to 5, with a narrow standard deviation of 0.7. The poorest job fit was reported by 0.3%, whereas optimal fit was reported by 21% of respondents. Job fit plotted against years in current practice had a logarithmic appearance typical of learning curves (Figure 1). An inflection point was visualized at around 2 years. For the purposes of this article, we refer to hospitalists' experience in the first 2 years of a job as an assimilation period, which is marked by a steep increase in job fit early when rapid learning or attrition took place. The years beyond the inflection point are characterized as an advancement period, when a more attenuated rise in job fit was experienced with time. The Spearman correlation between job fit and years in practice during the advancement period was 0.145 (n = 678, P < 0.001). Panel B of Table 1 displays the characteristics of respondents separately for the assimilation and advancement cohorts. Assimilation hospitalists in our sample had a mean age of 41.9 years and mean on‐the‐job experience of 4.3 years, reflecting that many hospitalists in the first 2 years of a job have made at least 1 job change in the past.

Figure 1

To show the effects of attrition and reselection, we first evaluated the proposition that hospitalists experience attrition (ie, intend to leave their jobs) in response to poor fit. Table 2 shows the correlations between job fit and the self‐reported intent to leave practice or reduce workload separately for the assimilation and advancement periods. For hospitalists in the assimilation period, job fit was negatively correlated with intent to leave current practice within 2 years and to leave hospital medicine within 5 years (P = 0.010 and 0.043, respectively). Hospitalists with <2 years in their current job, therefore, tended to consider attrition but not workload reduction to deal with poor job fit. On the other hand, hospitalists in the advancement period considered both attrition and workload reduction strategies in response to poor fit (all P < 0.001).

In Table 3, we further compared the median job fit across categories for the number of job switches. The median job fit during the assimilation period of hospitalists who had made 1 job change was slightly but statistically higher than the job fit of their counterparts who never left their first job (4.4 vs 4.0, P = 0.046). This suggests that job switching by hospitalists early in their jobs is associated with improved job fit (H1). However, the fit during the assimilation period of hospitalists who switched jobs twice or more was statistically no different from the fit of those in their first jobs, suggesting that the effect of the attrition‐reselection strategy is weak or inconsistent. The job fit for advancement period hospitalists was also different across the job change and no‐change categories. However, in the case of hospitalists later in their jobs, the median job fit was slightly but statistically lower among those who made job changes, revealing the potential drop in job fit that occurs when a hospitalist already established in his or her job starts over again in a new setting.

We hypothesized that hospitalists who achieved high job fit within a particular job were more likely to have engaged in activities that utilize a wider spectrum of their abilities. As shown in Table 4, hospitalists in the highest quartile of job fit were associated with a general trend toward higher odds of participating in a variety of common clinical and nonclinical hospitalist activities, but only the odds ratio associated with teaching achieved statistical significance (OR: 1.53, 95% CI: 1.01‐2.31) (H2).

Socialization with peers and the gradual sharing of values within organizations are hypothesized mechanisms for increasing job fit with time. We found that the number of years in current practice was positively correlated with job fit (Spearman coefficient R = 0.149, P < 0.001), organizational climate (R = 0.128, P < 0.001), and relationship with nonphysician staff (R = 0.102, P < 0.01). The association between years in practice and relationship with physician colleagues were weaker (R = 0.079, P < 0.05). Consistent with the episodic nature of patients' encounters with hospitalists, the measure of patient relationships was not significantly associated with length of time in job. In addition, we found substantial correlations among job fit, organizational climate, and all the relational measures (all R > 0.280, P < 0.001), indicating that hospitalists increasingly share the values of their organizations over time (H3).

Finally, we also hypothesized that poor job fit is associated with poor performance and quality of work life. Strong correlations with job fit were noted for stress (R = 0.307, P < 0.001), job burnout (R = 0.360, P < 0.001), and job satisfaction (R = 0.570, P < 0.001). Job fit (R = 0.147, P < 0.001), job burnout (R = 0.236, P < 0.001), stress (R = 0.305, P < 0.001), and job satisfaction (R = 0.224, P < 0.001) were all significantly correlated with the frequency of participating in suboptimal care (H4).

DISCUSSION

In this exploratory analysis, we validated in the hospitalist workforce several assumptions about person‐job fit that have been observed in workers of other industries. We observed attrition‐reselection (ie, job switching) as a strategy used by physicians to achieve better fit early in their job tenure, whereas job modification appeared to be more effective than attrition‐reselection among physicians already established in their jobs. We provided weak but plausible evidence that physicians with optimal job fit had a tendency to participate in activities (eg, teaching) that engage a wider set of interests and abilities. We also demonstrated the growth in hospitalists sharing the values of their organization through the time‐dependent associations among organizational climate, relational measures, and job fit. Finally, we found that physicians with suboptimal job fit were more likely to report poor performance in their work compared to those indicating optimal fit.

Our previous analysis of data from the Hospital Medicine Physician Worklife Survey exposed the widely variable work characteristics of hospitalist jobs in the US market and the equally variable preferences and priorities of individual hospitalists in selecting their work setting.[7] The implication of our present study is that hospitalists achieve the high levels of observed job fit using various strategies that aid their alignment with their employment. One of these strategies involves time, but physician longevity in practice may be both a determinant and product of good job fit. Although early job attrition may be necessary for fitting the right hospitalists to the right jobs, employers may appreciate the importance of retaining experienced hospitalists not only for cost and performance considerations but also for the growth of social capital in organizations consisting of enduring individuals. As our data suggest that hospitalists grow with their jobs, physicians may experience better fit with jobs that flexibly couple their work demands with benefits that address their individual work‐life needs over time. Another implication of this study is that job fit is a useful and predictive measure of job selection, performance, and retention. In light of studies that expose the limitations of job satisfaction as a measure influenced more by workers' dispositional affect (ie, their temperament and outlook) than their compatibility with their jobs,[28] job fit may add a functional dimension to traditional employee feedback measures.

There are limitations to this analysis. The most notable is the low survey response rate. Two reasons contributed to the fairly low rate of return. First, the original sampling frame included many outdated addresses and names of individuals who did not meet inclusion criteria. Although all sampled individuals who would have been excluded from the study could not be identified, we calculated our response rate without accounting for the proportion of potential ineligibles in the denominator population [Response Rate 2 (RR2) according to standards of the American Association of Public Opinion Research].[29] Second, the response rates of physician surveys have seen a steady decline over the years.[30] Respondents to our survey may be older and more experienced than US hospitalists in general. Although concerns about bias from under‐reporting cannot be fully addressed, we believe that the study sample is adequate for this preliminary study intended to translate the evidence of observed phenomena from the nonphysician to the physician workforces. The suboptimal response characteristics (high skew and low variability) of the generic person‐job fit survey scale used in this study indicate that a reliable survey instrument specifically designed to measure physician‐job fit need to be constructed de novo and validated for any future study. Although we performed simple analyses to support our assertions, few of our subanalyses may be underpowered, contributing to overinterpretation of the data. Additional empirical work is also necessary to assess the generalizability of this study's claims in other medical and surgical specialties. Such studies would also allow measurement of the sensitivity and specificity of physicians' self‐identification of poor job fit. Finally, additional investigations of this time‐dependent construct are more appropriately performed using a longitudinal study design to overcome the limitations inherent in this cross‐sectional analysis. Our conclusions about the time‐dependent features of job fit may be explained by other characteristics such as generational and cultural differences among hospitalists with varying experience.

As the US healthcare system reorganizes to bolster accountability,[31] we anticipate increasing interdependence between physicians and their employer organizations. Ultimately, the desired accountability in healthcare is likely to be obtained if physicians function not only as passive and interchangeable employees but as active stakeholders in the achievement of each organization's goals. A methodology for assessing the alignment of physicians and their jobs will continue to be important along the way.

Person‐organization fit concerns the conditions and consequences of compatibility between people and the organizations for which they work.[1] Studies of other industries have demonstrated that person‐organization fit informs the way individuals join, perform in, and are retained by organizations.[2] Person‐job fit is a closely related subordinate concept that concerns the alignment of workers and their job in as much as workers have needs that their job supplies, or conversely, jobs have requirements that certain workers' abilities can help meet.[3] Explorations of job fit in physicians and their work have recently emerged in a few investigations published in medical journals.[4, 5, 6, 7, 8] Further expanding the understanding of fit between physicians and their employment is important, because the decline of solo practices and recent emphasis on team‐based care have led to a growing number of US physicians working in organizations.[9]

The movement of physicians into employed situations may continue if certain types of Accountable Care Organizations take root.[10] And physicians may be primed to join employer organizations based on current career priorities of individuals in American society. Surveys of medical residents entering the workforce reveal more physicians preferring the security of being employees than starting their own practices.[11] Given these trends, job fit will inform our understanding of how personal and job characteristics facilitate recruitment, performance, satisfaction, and longevity of physician employees.

BACKGROUND

Virtually all hospitalists work in organizationshospitalsand are employees of hospitals, medical schools, physician group practices, or management companies, and therefore invariably function within organizational structures and systems.[7] In spite of their rapid growth in numbers, many employers have faced difficulties recruiting and retaining enough hospitalists to fill their staffing needs. Consequently, the US hospitalist workforce today is characterized by high salaries, work load, and attrition rates.[12]

In this evolving unsaturated market, the attraction‐selection‐attrition framework[13] provides a theoretical construct that predicts that hospitalists and their employers would seek congruence of goals and values early in their relationship through a process of trial and error. This framework assumes that early interactions between workers and organizations serve as opportunities for them to understand if job fit is poor and dissociate or remain affiliated as long as job fit is mutually acceptable. Therefore, job switching on average is expected to increase job fit because workers and organizations gain a better understanding of their own goals and values and choose more wisely the next time.

Other theoretical frameworks, such as the job characteristic model,[14] suggest that over time as workers stay at the same job, they continue to maintain and improve job fit through various workplace‐ or self‐modification strategies. For example, seniority status may have privileges (eg, less undesirable call), or workers may create privileged niches through the acquisition of new skills and abilities over time. Hospitalists' tendency to diversify their work‐related activities by incorporating administrative and teaching responsibilities[15] may thus contribute to improving job fit. Additionally, as a measure of complementarity among people who work together, job fit may be influenced by the quality of relationships among hospitalists and their coworkers through their reorientation to the prevailing organizational climate[16, 17] and increasing socialization.[18] Finally, given that experiential learning is known to contribute to better hospitalist work performance,[19] job fit may affect productivity and clinical outcomes vis‐‐vis quality of work life.

To test the validity of these assumptions in a sample of hospitalists, we critically appraised the following 4 hypotheses:

• Hypothesis 1 (H1): Job attrition and reselection improves job fit among hospitalists entering the job market.
• Hypothesis 2 (H2): Better job fit is achieved through hospitalists engaging a variety of personal skills and abilities.
• Hypothesis 3 (H3): Job fit increases with hospitalists' job duration together with socialization and internalization of organizational values.
• Hypothesis 4 (H4): Job fit is correlated with hospitalists' quality of work life.

METHODS

Analysis was performed on data from the 2009 to 2010 Hospital Medicine Physician Worklife Survey. The sample frame included nonmembers and members of Society of Hospital Medicine (SHM). Details about sampling strategy, data collection, and data quality are available in previous publications.[7, 20] The 118‐item survey instrument, including 9 demographic items and 24 practice and job characteristic items, was administered by mail. Examples of information solicited through these items included respondents' practice model, the number of hospitalist jobs they have held, and the specific kinds of clinical and nonclinical activities they performed as part of their current job.

We used a reliable but broad and generic measure of self‐perceived person‐job fit.[21] The survey items of the 5‐point Likert‐type scale anchored between strongly disagree and strongly agree were: I feel that my work utilizes my full abilities, I feel competent and fully able to handle my job, my job gives me the chance to do the things I feel I do best, I feel that my job and I are well‐matched, I feel I have adequate preparation for the job I now hold. The quality of hospitalists' relationships with physician colleagues, staff, and patients as well as job satisfaction was measured using scales adapted from the Physician Worklife Study.[22] Organizational climate was measured using an adapted scale from the Minimizing Error, Maximizing Outcome study incorporating 3 items from the cohesiveness subscale, 4 items from the organizational trust subscale, and 1 item from the quality emphasis subscale that were most pertinent to hospitalists' relationship with their organizations.[23] Intent to leave practice or reduce work hours was measured using 5 items from the Multi‐Center Hospitalist Survey Project.[24] Frequency of participation in suboptimal patient care was measured by adapting 3 items from the suboptimal reported practice subscale and 2 items from the suboptimal patient care subscale developed by Shanafelt et al.[25] Stress and job burnout were assessed using validated measures.[26, 27] Detailed descriptions of the response rate calculation and imputation of missing item data are available in previous publications.[7, 20]

Mean, variance, range, and skew were used to characterize the responses to the job fit survey scale. A table of respondent characteristics was constructed. A visual representation of job fit by individual hospitalist year in current practice was created, first, by plotting a locally weighted scatterplot smoothing curve to examine the shape of the general relationship, and second, by fitting a similarly contoured functional polynomial curve with 95% confidence intervals (CI) to a plot of the mean and interquartile range of job fit for each year in current practice. Spearman partial correlations were calculated for job fit and each of the 5 items addressing likelihood of leaving practice or reducing workload adjusted for gender to control for the higher proportion of women who plan to work part time. Median (interquartile range) job fit was calculated for categories defined by the number of job changes and compared with the reference category (no job change) using the nonparametric rank sum test for comparing non‐normally distributed data. Multivariate logistic regression models were used to calculate the odds ratio (OR) of participating in each of several clinical and nonclinical hospitalist activities between respondents whose job fit score was optimal (5 on a 5‐point scale) and less than optimal controlling for covariates that influence the likelihood of participating in these activities (years in current practice, practice model, and specialty training). A Spearman correlation matrix was created to assess interscale correlations among organizational parameters (years in current practice, job fit, organizational climate, and relationship with colleagues, staff, and patients). Finally, a separate Spearman correlation matrix was created to assess the interscale correlations among individual worker parameters (job fit, suboptimal patient care, job burnout, stress, and job satisfaction). Statistical significance was defined as P value <0.05, and all analyses were performed on Stata 11.0 (StataCorp, College Station, TX). The Northwestern University institutional review board approved this study.

RESULTS

Respondents included 816 hospitalists belonging to around 700 unique organizations. The adjusted response rate from the stratified sample was 26%. Respondents and nonrespondents were similar with regard to geographic region and model of practice, but respondents were more likely to be members of the SHM than nonrespondents. Panel A of Table 1 shows the demographic characteristics of the respondents. The mean age was 44.3 years, and about one‐third were women. The average hospitalist had about 7 years of experience in the specialty and about 5 years with their current hospitalist job. The majority were trained in internal medicine or one of its subspecialties, whereas pediatricians, family physicians, and physicians with other training made up the remainder.

Job fit was highly skewed toward optimum fit, with a mean of 4.3 on a scale of 1 to 5, with a narrow standard deviation of 0.7. The poorest job fit was reported by 0.3%, whereas optimal fit was reported by 21% of respondents. Job fit plotted against years in current practice had a logarithmic appearance typical of learning curves (Figure 1). An inflection point was visualized at around 2 years. For the purposes of this article, we refer to hospitalists' experience in the first 2 years of a job as an assimilation period, which is marked by a steep increase in job fit early when rapid learning or attrition took place. The years beyond the inflection point are characterized as an advancement period, when a more attenuated rise in job fit was experienced with time. The Spearman correlation between job fit and years in practice during the advancement period was 0.145 (n = 678, P < 0.001). Panel B of Table 1 displays the characteristics of respondents separately for the assimilation and advancement cohorts. Assimilation hospitalists in our sample had a mean age of 41.9 years and mean on‐the‐job experience of 4.3 years, reflecting that many hospitalists in the first 2 years of a job have made at least 1 job change in the past.

Figure 1

To show the effects of attrition and reselection, we first evaluated the proposition that hospitalists experience attrition (ie, intend to leave their jobs) in response to poor fit. Table 2 shows the correlations between job fit and the self‐reported intent to leave practice or reduce workload separately for the assimilation and advancement periods. For hospitalists in the assimilation period, job fit was negatively correlated with intent to leave current practice within 2 years and to leave hospital medicine within 5 years (P = 0.010 and 0.043, respectively). Hospitalists with <2 years in their current job, therefore, tended to consider attrition but not workload reduction to deal with poor job fit. On the other hand, hospitalists in the advancement period considered both attrition and workload reduction strategies in response to poor fit (all P < 0.001).

In Table 3, we further compared the median job fit across categories for the number of job switches. The median job fit during the assimilation period of hospitalists who had made 1 job change was slightly but statistically higher than the job fit of their counterparts who never left their first job (4.4 vs 4.0, P = 0.046). This suggests that job switching by hospitalists early in their jobs is associated with improved job fit (H1). However, the fit during the assimilation period of hospitalists who switched jobs twice or more was statistically no different from the fit of those in their first jobs, suggesting that the effect of the attrition‐reselection strategy is weak or inconsistent. The job fit for advancement period hospitalists was also different across the job change and no‐change categories. However, in the case of hospitalists later in their jobs, the median job fit was slightly but statistically lower among those who made job changes, revealing the potential drop in job fit that occurs when a hospitalist already established in his or her job starts over again in a new setting.

We hypothesized that hospitalists who achieved high job fit within a particular job were more likely to have engaged in activities that utilize a wider spectrum of their abilities. As shown in Table 4, hospitalists in the highest quartile of job fit were associated with a general trend toward higher odds of participating in a variety of common clinical and nonclinical hospitalist activities, but only the odds ratio associated with teaching achieved statistical significance (OR: 1.53, 95% CI: 1.01‐2.31) (H2).

Socialization with peers and the gradual sharing of values within organizations are hypothesized mechanisms for increasing job fit with time. We found that the number of years in current practice was positively correlated with job fit (Spearman coefficient R = 0.149, P < 0.001), organizational climate (R = 0.128, P < 0.001), and relationship with nonphysician staff (R = 0.102, P < 0.01). The association between years in practice and relationship with physician colleagues were weaker (R = 0.079, P < 0.05). Consistent with the episodic nature of patients' encounters with hospitalists, the measure of patient relationships was not significantly associated with length of time in job. In addition, we found substantial correlations among job fit, organizational climate, and all the relational measures (all R > 0.280, P < 0.001), indicating that hospitalists increasingly share the values of their organizations over time (H3).

Finally, we also hypothesized that poor job fit is associated with poor performance and quality of work life. Strong correlations with job fit were noted for stress (R = 0.307, P < 0.001), job burnout (R = 0.360, P < 0.001), and job satisfaction (R = 0.570, P < 0.001). Job fit (R = 0.147, P < 0.001), job burnout (R = 0.236, P < 0.001), stress (R = 0.305, P < 0.001), and job satisfaction (R = 0.224, P < 0.001) were all significantly correlated with the frequency of participating in suboptimal care (H4).

DISCUSSION

In this exploratory analysis, we validated in the hospitalist workforce several assumptions about person‐job fit that have been observed in workers of other industries. We observed attrition‐reselection (ie, job switching) as a strategy used by physicians to achieve better fit early in their job tenure, whereas job modification appeared to be more effective than attrition‐reselection among physicians already established in their jobs. We provided weak but plausible evidence that physicians with optimal job fit had a tendency to participate in activities (eg, teaching) that engage a wider set of interests and abilities. We also demonstrated the growth in hospitalists sharing the values of their organization through the time‐dependent associations among organizational climate, relational measures, and job fit. Finally, we found that physicians with suboptimal job fit were more likely to report poor performance in their work compared to those indicating optimal fit.

Our previous analysis of data from the Hospital Medicine Physician Worklife Survey exposed the widely variable work characteristics of hospitalist jobs in the US market and the equally variable preferences and priorities of individual hospitalists in selecting their work setting.[7] The implication of our present study is that hospitalists achieve the high levels of observed job fit using various strategies that aid their alignment with their employment. One of these strategies involves time, but physician longevity in practice may be both a determinant and product of good job fit. Although early job attrition may be necessary for fitting the right hospitalists to the right jobs, employers may appreciate the importance of retaining experienced hospitalists not only for cost and performance considerations but also for the growth of social capital in organizations consisting of enduring individuals. As our data suggest that hospitalists grow with their jobs, physicians may experience better fit with jobs that flexibly couple their work demands with benefits that address their individual work‐life needs over time. Another implication of this study is that job fit is a useful and predictive measure of job selection, performance, and retention. In light of studies that expose the limitations of job satisfaction as a measure influenced more by workers' dispositional affect (ie, their temperament and outlook) than their compatibility with their jobs,[28] job fit may add a functional dimension to traditional employee feedback measures.

There are limitations to this analysis. The most notable is the low survey response rate. Two reasons contributed to the fairly low rate of return. First, the original sampling frame included many outdated addresses and names of individuals who did not meet inclusion criteria. Although all sampled individuals who would have been excluded from the study could not be identified, we calculated our response rate without accounting for the proportion of potential ineligibles in the denominator population [Response Rate 2 (RR2) according to standards of the American Association of Public Opinion Research].[29] Second, the response rates of physician surveys have seen a steady decline over the years.[30] Respondents to our survey may be older and more experienced than US hospitalists in general. Although concerns about bias from under‐reporting cannot be fully addressed, we believe that the study sample is adequate for this preliminary study intended to translate the evidence of observed phenomena from the nonphysician to the physician workforces. The suboptimal response characteristics (high skew and low variability) of the generic person‐job fit survey scale used in this study indicate that a reliable survey instrument specifically designed to measure physician‐job fit need to be constructed de novo and validated for any future study. Although we performed simple analyses to support our assertions, few of our subanalyses may be underpowered, contributing to overinterpretation of the data. Additional empirical work is also necessary to assess the generalizability of this study's claims in other medical and surgical specialties. Such studies would also allow measurement of the sensitivity and specificity of physicians' self‐identification of poor job fit. Finally, additional investigations of this time‐dependent construct are more appropriately performed using a longitudinal study design to overcome the limitations inherent in this cross‐sectional analysis. Our conclusions about the time‐dependent features of job fit may be explained by other characteristics such as generational and cultural differences among hospitalists with varying experience.

As the US healthcare system reorganizes to bolster accountability,[31] we anticipate increasing interdependence between physicians and their employer organizations. Ultimately, the desired accountability in healthcare is likely to be obtained if physicians function not only as passive and interchangeable employees but as active stakeholders in the achievement of each organization's goals. A methodology for assessing the alignment of physicians and their jobs will continue to be important along the way.

Person‐organization fit concerns the conditions and consequences of compatibility between people and the organizations for which they work.[1] Studies of other industries have demonstrated that person‐organization fit informs the way individuals join, perform in, and are retained by organizations.[2] Person‐job fit is a closely related subordinate concept that concerns the alignment of workers and their job in as much as workers have needs that their job supplies, or conversely, jobs have requirements that certain workers' abilities can help meet.[3] Explorations of job fit in physicians and their work have recently emerged in a few investigations published in medical journals.[4, 5, 6, 7, 8] Further expanding the understanding of fit between physicians and their employment is important, because the decline of solo practices and recent emphasis on team‐based care have led to a growing number of US physicians working in organizations.[9]

The movement of physicians into employed situations may continue if certain types of Accountable Care Organizations take root.[10] And physicians may be primed to join employer organizations based on current career priorities of individuals in American society. Surveys of medical residents entering the workforce reveal more physicians preferring the security of being employees than starting their own practices.[11] Given these trends, job fit will inform our understanding of how personal and job characteristics facilitate recruitment, performance, satisfaction, and longevity of physician employees.

BACKGROUND

Virtually all hospitalists work in organizationshospitalsand are employees of hospitals, medical schools, physician group practices, or management companies, and therefore invariably function within organizational structures and systems.[7] In spite of their rapid growth in numbers, many employers have faced difficulties recruiting and retaining enough hospitalists to fill their staffing needs. Consequently, the US hospitalist workforce today is characterized by high salaries, work load, and attrition rates.[12]

In this evolving unsaturated market, the attraction‐selection‐attrition framework[13] provides a theoretical construct that predicts that hospitalists and their employers would seek congruence of goals and values early in their relationship through a process of trial and error. This framework assumes that early interactions between workers and organizations serve as opportunities for them to understand if job fit is poor and dissociate or remain affiliated as long as job fit is mutually acceptable. Therefore, job switching on average is expected to increase job fit because workers and organizations gain a better understanding of their own goals and values and choose more wisely the next time.

Other theoretical frameworks, such as the job characteristic model,[14] suggest that over time as workers stay at the same job, they continue to maintain and improve job fit through various workplace‐ or self‐modification strategies. For example, seniority status may have privileges (eg, less undesirable call), or workers may create privileged niches through the acquisition of new skills and abilities over time. Hospitalists' tendency to diversify their work‐related activities by incorporating administrative and teaching responsibilities[15] may thus contribute to improving job fit. Additionally, as a measure of complementarity among people who work together, job fit may be influenced by the quality of relationships among hospitalists and their coworkers through their reorientation to the prevailing organizational climate[16, 17] and increasing socialization.[18] Finally, given that experiential learning is known to contribute to better hospitalist work performance,[19] job fit may affect productivity and clinical outcomes vis‐‐vis quality of work life.

To test the validity of these assumptions in a sample of hospitalists, we critically appraised the following 4 hypotheses:

• Hypothesis 1 (H1): Job attrition and reselection improves job fit among hospitalists entering the job market.
• Hypothesis 2 (H2): Better job fit is achieved through hospitalists engaging a variety of personal skills and abilities.
• Hypothesis 3 (H3): Job fit increases with hospitalists' job duration together with socialization and internalization of organizational values.
• Hypothesis 4 (H4): Job fit is correlated with hospitalists' quality of work life.

METHODS

Analysis was performed on data from the 2009 to 2010 Hospital Medicine Physician Worklife Survey. The sample frame included nonmembers and members of Society of Hospital Medicine (SHM). Details about sampling strategy, data collection, and data quality are available in previous publications.[7, 20] The 118‐item survey instrument, including 9 demographic items and 24 practice and job characteristic items, was administered by mail. Examples of information solicited through these items included respondents' practice model, the number of hospitalist jobs they have held, and the specific kinds of clinical and nonclinical activities they performed as part of their current job.

We used a reliable but broad and generic measure of self‐perceived person‐job fit.[21] The survey items of the 5‐point Likert‐type scale anchored between strongly disagree and strongly agree were: I feel that my work utilizes my full abilities, I feel competent and fully able to handle my job, my job gives me the chance to do the things I feel I do best, I feel that my job and I are well‐matched, I feel I have adequate preparation for the job I now hold. The quality of hospitalists' relationships with physician colleagues, staff, and patients as well as job satisfaction was measured using scales adapted from the Physician Worklife Study.[22] Organizational climate was measured using an adapted scale from the Minimizing Error, Maximizing Outcome study incorporating 3 items from the cohesiveness subscale, 4 items from the organizational trust subscale, and 1 item from the quality emphasis subscale that were most pertinent to hospitalists' relationship with their organizations.[23] Intent to leave practice or reduce work hours was measured using 5 items from the Multi‐Center Hospitalist Survey Project.[24] Frequency of participation in suboptimal patient care was measured by adapting 3 items from the suboptimal reported practice subscale and 2 items from the suboptimal patient care subscale developed by Shanafelt et al.[25] Stress and job burnout were assessed using validated measures.[26, 27] Detailed descriptions of the response rate calculation and imputation of missing item data are available in previous publications.[7, 20]

Mean, variance, range, and skew were used to characterize the responses to the job fit survey scale. A table of respondent characteristics was constructed. A visual representation of job fit by individual hospitalist year in current practice was created, first, by plotting a locally weighted scatterplot smoothing curve to examine the shape of the general relationship, and second, by fitting a similarly contoured functional polynomial curve with 95% confidence intervals (CI) to a plot of the mean and interquartile range of job fit for each year in current practice. Spearman partial correlations were calculated for job fit and each of the 5 items addressing likelihood of leaving practice or reducing workload adjusted for gender to control for the higher proportion of women who plan to work part time. Median (interquartile range) job fit was calculated for categories defined by the number of job changes and compared with the reference category (no job change) using the nonparametric rank sum test for comparing non‐normally distributed data. Multivariate logistic regression models were used to calculate the odds ratio (OR) of participating in each of several clinical and nonclinical hospitalist activities between respondents whose job fit score was optimal (5 on a 5‐point scale) and less than optimal controlling for covariates that influence the likelihood of participating in these activities (years in current practice, practice model, and specialty training). A Spearman correlation matrix was created to assess interscale correlations among organizational parameters (years in current practice, job fit, organizational climate, and relationship with colleagues, staff, and patients). Finally, a separate Spearman correlation matrix was created to assess the interscale correlations among individual worker parameters (job fit, suboptimal patient care, job burnout, stress, and job satisfaction). Statistical significance was defined as P value <0.05, and all analyses were performed on Stata 11.0 (StataCorp, College Station, TX). The Northwestern University institutional review board approved this study.

RESULTS

Respondents included 816 hospitalists belonging to around 700 unique organizations. The adjusted response rate from the stratified sample was 26%. Respondents and nonrespondents were similar with regard to geographic region and model of practice, but respondents were more likely to be members of the SHM than nonrespondents. Panel A of Table 1 shows the demographic characteristics of the respondents. The mean age was 44.3 years, and about one‐third were women. The average hospitalist had about 7 years of experience in the specialty and about 5 years with their current hospitalist job. The majority were trained in internal medicine or one of its subspecialties, whereas pediatricians, family physicians, and physicians with other training made up the remainder.

Job fit was highly skewed toward optimum fit, with a mean of 4.3 on a scale of 1 to 5, with a narrow standard deviation of 0.7. The poorest job fit was reported by 0.3%, whereas optimal fit was reported by 21% of respondents. Job fit plotted against years in current practice had a logarithmic appearance typical of learning curves (Figure 1). An inflection point was visualized at around 2 years. For the purposes of this article, we refer to hospitalists' experience in the first 2 years of a job as an assimilation period, which is marked by a steep increase in job fit early when rapid learning or attrition took place. The years beyond the inflection point are characterized as an advancement period, when a more attenuated rise in job fit was experienced with time. The Spearman correlation between job fit and years in practice during the advancement period was 0.145 (n = 678, P < 0.001). Panel B of Table 1 displays the characteristics of respondents separately for the assimilation and advancement cohorts. Assimilation hospitalists in our sample had a mean age of 41.9 years and mean on‐the‐job experience of 4.3 years, reflecting that many hospitalists in the first 2 years of a job have made at least 1 job change in the past.

Figure 1

To show the effects of attrition and reselection, we first evaluated the proposition that hospitalists experience attrition (ie, intend to leave their jobs) in response to poor fit. Table 2 shows the correlations between job fit and the self‐reported intent to leave practice or reduce workload separately for the assimilation and advancement periods. For hospitalists in the assimilation period, job fit was negatively correlated with intent to leave current practice within 2 years and to leave hospital medicine within 5 years (P = 0.010 and 0.043, respectively). Hospitalists with <2 years in their current job, therefore, tended to consider attrition but not workload reduction to deal with poor job fit. On the other hand, hospitalists in the advancement period considered both attrition and workload reduction strategies in response to poor fit (all P < 0.001).

In Table 3, we further compared the median job fit across categories for the number of job switches. The median job fit during the assimilation period of hospitalists who had made 1 job change was slightly but statistically higher than the job fit of their counterparts who never left their first job (4.4 vs 4.0, P = 0.046). This suggests that job switching by hospitalists early in their jobs is associated with improved job fit (H1). However, the fit during the assimilation period of hospitalists who switched jobs twice or more was statistically no different from the fit of those in their first jobs, suggesting that the effect of the attrition‐reselection strategy is weak or inconsistent. The job fit for advancement period hospitalists was also different across the job change and no‐change categories. However, in the case of hospitalists later in their jobs, the median job fit was slightly but statistically lower among those who made job changes, revealing the potential drop in job fit that occurs when a hospitalist already established in his or her job starts over again in a new setting.

We hypothesized that hospitalists who achieved high job fit within a particular job were more likely to have engaged in activities that utilize a wider spectrum of their abilities. As shown in Table 4, hospitalists in the highest quartile of job fit were associated with a general trend toward higher odds of participating in a variety of common clinical and nonclinical hospitalist activities, but only the odds ratio associated with teaching achieved statistical significance (OR: 1.53, 95% CI: 1.01‐2.31) (H2).

Socialization with peers and the gradual sharing of values within organizations are hypothesized mechanisms for increasing job fit with time. We found that the number of years in current practice was positively correlated with job fit (Spearman coefficient R = 0.149, P < 0.001), organizational climate (R = 0.128, P < 0.001), and relationship with nonphysician staff (R = 0.102, P < 0.01). The association between years in practice and relationship with physician colleagues were weaker (R = 0.079, P < 0.05). Consistent with the episodic nature of patients' encounters with hospitalists, the measure of patient relationships was not significantly associated with length of time in job. In addition, we found substantial correlations among job fit, organizational climate, and all the relational measures (all R > 0.280, P < 0.001), indicating that hospitalists increasingly share the values of their organizations over time (H3).

Finally, we also hypothesized that poor job fit is associated with poor performance and quality of work life. Strong correlations with job fit were noted for stress (R = 0.307, P < 0.001), job burnout (R = 0.360, P < 0.001), and job satisfaction (R = 0.570, P < 0.001). Job fit (R = 0.147, P < 0.001), job burnout (R = 0.236, P < 0.001), stress (R = 0.305, P < 0.001), and job satisfaction (R = 0.224, P < 0.001) were all significantly correlated with the frequency of participating in suboptimal care (H4).

DISCUSSION

In this exploratory analysis, we validated in the hospitalist workforce several assumptions about person‐job fit that have been observed in workers of other industries. We observed attrition‐reselection (ie, job switching) as a strategy used by physicians to achieve better fit early in their job tenure, whereas job modification appeared to be more effective than attrition‐reselection among physicians already established in their jobs. We provided weak but plausible evidence that physicians with optimal job fit had a tendency to participate in activities (eg, teaching) that engage a wider set of interests and abilities. We also demonstrated the growth in hospitalists sharing the values of their organization through the time‐dependent associations among organizational climate, relational measures, and job fit. Finally, we found that physicians with suboptimal job fit were more likely to report poor performance in their work compared to those indicating optimal fit.

Our previous analysis of data from the Hospital Medicine Physician Worklife Survey exposed the widely variable work characteristics of hospitalist jobs in the US market and the equally variable preferences and priorities of individual hospitalists in selecting their work setting.[7] The implication of our present study is that hospitalists achieve the high levels of observed job fit using various strategies that aid their alignment with their employment. One of these strategies involves time, but physician longevity in practice may be both a determinant and product of good job fit. Although early job attrition may be necessary for fitting the right hospitalists to the right jobs, employers may appreciate the importance of retaining experienced hospitalists not only for cost and performance considerations but also for the growth of social capital in organizations consisting of enduring individuals. As our data suggest that hospitalists grow with their jobs, physicians may experience better fit with jobs that flexibly couple their work demands with benefits that address their individual work‐life needs over time. Another implication of this study is that job fit is a useful and predictive measure of job selection, performance, and retention. In light of studies that expose the limitations of job satisfaction as a measure influenced more by workers' dispositional affect (ie, their temperament and outlook) than their compatibility with their jobs,[28] job fit may add a functional dimension to traditional employee feedback measures.

There are limitations to this analysis. The most notable is the low survey response rate. Two reasons contributed to the fairly low rate of return. First, the original sampling frame included many outdated addresses and names of individuals who did not meet inclusion criteria. Although all sampled individuals who would have been excluded from the study could not be identified, we calculated our response rate without accounting for the proportion of potential ineligibles in the denominator population [Response Rate 2 (RR2) according to standards of the American Association of Public Opinion Research].[29] Second, the response rates of physician surveys have seen a steady decline over the years.[30] Respondents to our survey may be older and more experienced than US hospitalists in general. Although concerns about bias from under‐reporting cannot be fully addressed, we believe that the study sample is adequate for this preliminary study intended to translate the evidence of observed phenomena from the nonphysician to the physician workforces. The suboptimal response characteristics (high skew and low variability) of the generic person‐job fit survey scale used in this study indicate that a reliable survey instrument specifically designed to measure physician‐job fit need to be constructed de novo and validated for any future study. Although we performed simple analyses to support our assertions, few of our subanalyses may be underpowered, contributing to overinterpretation of the data. Additional empirical work is also necessary to assess the generalizability of this study's claims in other medical and surgical specialties. Such studies would also allow measurement of the sensitivity and specificity of physicians' self‐identification of poor job fit. Finally, additional investigations of this time‐dependent construct are more appropriately performed using a longitudinal study design to overcome the limitations inherent in this cross‐sectional analysis. Our conclusions about the time‐dependent features of job fit may be explained by other characteristics such as generational and cultural differences among hospitalists with varying experience.

As the US healthcare system reorganizes to bolster accountability,[31] we anticipate increasing interdependence between physicians and their employer organizations. Ultimately, the desired accountability in healthcare is likely to be obtained if physicians function not only as passive and interchangeable employees but as active stakeholders in the achievement of each organization's goals. A methodology for assessing the alignment of physicians and their jobs will continue to be important along the way.