It’s elementary: Watson aids in breast cancer decisions

SAN ANTONIO – When oncologists at Manipal Hospitals in Bangalore, India, need help with a breast cancer conundrum, they can make the electronic equivalent of the famous cry for help “Watson, come here, I want you!”

In this case, Watson is not the real-life sidekick of Alexander Graham Bell or the fictional companion of Sherlock Holmes, but Watson for Oncology (WFO), an IBM-created artificial intelligence platform being developed in the United States and used in clinical practice in India to help guide clinical decision making but not to replace clinicians’ judgment, explained S.P. Somashekhar, MBBS, MS, MCH, FRCS, chairman of the Manipal Comprehensive Cancer Center.

“Watson in oncology is something which understands natural language, adapts and learns, and it [creates a] hypothesis based on the best available evidence on the best treatment for patients,” he said in a briefing at the San Antonio Breast Cancer Symposium.

This is no game

WFO is the clinical cousin of the artificial intelligence platform, also named Watson, that beat all-time champion Ken Jennings on the television game show “Jeopardy!” The system was named after Thomas J. Watson, IBM’s first chief executive officer.

The version of Watson used in India was developed by physicians and investigators at Memorial Sloan Kettering Cancer Center (MSKCC) in New York and by IBM. Watson for Oncology is fed national treatment guidelines, more than 1,000 training cases, MSKCC internal guidelines, and medical literature curated by MSKCC; it then chews over the data and spits out evidence-based recommendations.

At Manipal Hospitals, the electronic medical record contains an “Ask Watson” button that allows clinicians to interrogate the artificial entity’s vast stores of medical data to offer recommendations that clinicians can then use, modify, or reject at their discretion.

The system analyzes more than 100 patient attributes and then offers options with a green label for “recommended,” amber label “for consideration,” or a red label for “not recommended.”

Concordance study

At SABCS, Dr. Somashekhar presented results of a study evaluating the concordance of treatment recommendations between WFO and the members of the Manipal Multidisciplinary Tumor Board (MTB).

They looked at data on 638 patients treated in their hospital system over the three prior years, assessing the MTBs initial, best joint decision at the time of the original treatment decision (T1), and compared it with WFO’s recommendation made in 2016 (T2) and with a blinded MTB re-review of nonconcordant cases, also made in 2016.

Evaluating concordance by stage, they found that, for 514 cases of nonmetastatic disease, there was 79% concordance between the original MTB decisions and WFO’s recommendations for therapies in the combined “for consideration” and “recommended” categories.

For 124 cases of metastatic disease, however, Watson and its human counterparts were more frequently at odds, with only a 46% concordance.

In a subset analysis by receptor status, the investigators found that Watson was best – that is, most highly concordant – in triple-negative disease, with a 67.9% concordance in regard to MTB choices. In contrast, for patients with metastatic disease negative for the human epidermal growth factor receptor-2 (HER2), the concordance rate was a low 35%.

Dr. Somashekhar commented that part of the explanation for the discordant concordance rates by tumor subtype can be attributed to the fact that patients with triple-negative breast cancers have fewer treatment options than patients with HER2 negative–only tumors, who have a much broader array of possibilities.

The investigators also saw an overall concordance of WFO with the MTB recommendation at the time of the initial decision of 73%. But following 2016 MTB re-review of the original nonconcordant cases, the concordance between machine and mankind increased to 90%.

One area where WFO has its two-legged colleagues beaten hands down, however, is in the time it takes to capture and analyze data: humans took a mean of 20 minutes, Watson a median of 40 seconds.

The investigators acknowledged that WFO represents a further step toward personalized medicine, but emphasized that software will never replace the doctor-patient relationship or override the treating physician’s decisions.

‘This one would actually help’

At the briefing, moderator C. Kent Osborne, MD, from Baylor College of Medicine, Houston, commented, “I guess this not going to put us out of business as physicians – at least I hope not.”

Asked whether WFO could be a useful clinical tool or just another nuisance task added to an already overcrowded clinic schedule, Dr. Osborne responded, “I think this one would actually help.”

Anne Blaes, MD, from the University of Minnesota, Minneapolis, who also presented data at the briefing but was not involved in the concordance study, commented that, for oncologists practicing in the community, WFO offers the opportunity to get easy access to comprehensive, high-level information and expert opinions usually available only to those who toil in academic medical centers.

This study was investigative and received no external funding. Dr. Somashekhar, Dr. Osborne, and Dr. Blaes reported no conflicts of interest.

SAN ANTONIO – When oncologists at Manipal Hospitals in Bangalore, India, need help with a breast cancer conundrum, they can make the electronic equivalent of the famous cry for help “Watson, come here, I want you!”

In this case, Watson is not the real-life sidekick of Alexander Graham Bell or the fictional companion of Sherlock Holmes, but Watson for Oncology (WFO), an IBM-created artificial intelligence platform being developed in the United States and used in clinical practice in India to help guide clinical decision making but not to replace clinicians’ judgment, explained S.P. Somashekhar, MBBS, MS, MCH, FRCS, chairman of the Manipal Comprehensive Cancer Center.

“Watson in oncology is something which understands natural language, adapts and learns, and it [creates a] hypothesis based on the best available evidence on the best treatment for patients,” he said in a briefing at the San Antonio Breast Cancer Symposium.

This is no game

WFO is the clinical cousin of the artificial intelligence platform, also named Watson, that beat all-time champion Ken Jennings on the television game show “Jeopardy!” The system was named after Thomas J. Watson, IBM’s first chief executive officer.

The version of Watson used in India was developed by physicians and investigators at Memorial Sloan Kettering Cancer Center (MSKCC) in New York and by IBM. Watson for Oncology is fed national treatment guidelines, more than 1,000 training cases, MSKCC internal guidelines, and medical literature curated by MSKCC; it then chews over the data and spits out evidence-based recommendations.

At Manipal Hospitals, the electronic medical record contains an “Ask Watson” button that allows clinicians to interrogate the artificial entity’s vast stores of medical data to offer recommendations that clinicians can then use, modify, or reject at their discretion.

The system analyzes more than 100 patient attributes and then offers options with a green label for “recommended,” amber label “for consideration,” or a red label for “not recommended.”

Concordance study

At SABCS, Dr. Somashekhar presented results of a study evaluating the concordance of treatment recommendations between WFO and the members of the Manipal Multidisciplinary Tumor Board (MTB).

They looked at data on 638 patients treated in their hospital system over the three prior years, assessing the MTBs initial, best joint decision at the time of the original treatment decision (T1), and compared it with WFO’s recommendation made in 2016 (T2) and with a blinded MTB re-review of nonconcordant cases, also made in 2016.

Evaluating concordance by stage, they found that, for 514 cases of nonmetastatic disease, there was 79% concordance between the original MTB decisions and WFO’s recommendations for therapies in the combined “for consideration” and “recommended” categories.

For 124 cases of metastatic disease, however, Watson and its human counterparts were more frequently at odds, with only a 46% concordance.

In a subset analysis by receptor status, the investigators found that Watson was best – that is, most highly concordant – in triple-negative disease, with a 67.9% concordance in regard to MTB choices. In contrast, for patients with metastatic disease negative for the human epidermal growth factor receptor-2 (HER2), the concordance rate was a low 35%.

Dr. Somashekhar commented that part of the explanation for the discordant concordance rates by tumor subtype can be attributed to the fact that patients with triple-negative breast cancers have fewer treatment options than patients with HER2 negative–only tumors, who have a much broader array of possibilities.

The investigators also saw an overall concordance of WFO with the MTB recommendation at the time of the initial decision of 73%. But following 2016 MTB re-review of the original nonconcordant cases, the concordance between machine and mankind increased to 90%.

One area where WFO has its two-legged colleagues beaten hands down, however, is in the time it takes to capture and analyze data: humans took a mean of 20 minutes, Watson a median of 40 seconds.

The investigators acknowledged that WFO represents a further step toward personalized medicine, but emphasized that software will never replace the doctor-patient relationship or override the treating physician’s decisions.

‘This one would actually help’

At the briefing, moderator C. Kent Osborne, MD, from Baylor College of Medicine, Houston, commented, “I guess this not going to put us out of business as physicians – at least I hope not.”

Asked whether WFO could be a useful clinical tool or just another nuisance task added to an already overcrowded clinic schedule, Dr. Osborne responded, “I think this one would actually help.”

Anne Blaes, MD, from the University of Minnesota, Minneapolis, who also presented data at the briefing but was not involved in the concordance study, commented that, for oncologists practicing in the community, WFO offers the opportunity to get easy access to comprehensive, high-level information and expert opinions usually available only to those who toil in academic medical centers.

This study was investigative and received no external funding. Dr. Somashekhar, Dr. Osborne, and Dr. Blaes reported no conflicts of interest.

SAN ANTONIO – When oncologists at Manipal Hospitals in Bangalore, India, need help with a breast cancer conundrum, they can make the electronic equivalent of the famous cry for help “Watson, come here, I want you!”

In this case, Watson is not the real-life sidekick of Alexander Graham Bell or the fictional companion of Sherlock Holmes, but Watson for Oncology (WFO), an IBM-created artificial intelligence platform being developed in the United States and used in clinical practice in India to help guide clinical decision making but not to replace clinicians’ judgment, explained S.P. Somashekhar, MBBS, MS, MCH, FRCS, chairman of the Manipal Comprehensive Cancer Center.

“Watson in oncology is something which understands natural language, adapts and learns, and it [creates a] hypothesis based on the best available evidence on the best treatment for patients,” he said in a briefing at the San Antonio Breast Cancer Symposium.

This is no game

WFO is the clinical cousin of the artificial intelligence platform, also named Watson, that beat all-time champion Ken Jennings on the television game show “Jeopardy!” The system was named after Thomas J. Watson, IBM’s first chief executive officer.

The version of Watson used in India was developed by physicians and investigators at Memorial Sloan Kettering Cancer Center (MSKCC) in New York and by IBM. Watson for Oncology is fed national treatment guidelines, more than 1,000 training cases, MSKCC internal guidelines, and medical literature curated by MSKCC; it then chews over the data and spits out evidence-based recommendations.

At Manipal Hospitals, the electronic medical record contains an “Ask Watson” button that allows clinicians to interrogate the artificial entity’s vast stores of medical data to offer recommendations that clinicians can then use, modify, or reject at their discretion.

The system analyzes more than 100 patient attributes and then offers options with a green label for “recommended,” amber label “for consideration,” or a red label for “not recommended.”

Concordance study

At SABCS, Dr. Somashekhar presented results of a study evaluating the concordance of treatment recommendations between WFO and the members of the Manipal Multidisciplinary Tumor Board (MTB).

They looked at data on 638 patients treated in their hospital system over the three prior years, assessing the MTBs initial, best joint decision at the time of the original treatment decision (T1), and compared it with WFO’s recommendation made in 2016 (T2) and with a blinded MTB re-review of nonconcordant cases, also made in 2016.

Evaluating concordance by stage, they found that, for 514 cases of nonmetastatic disease, there was 79% concordance between the original MTB decisions and WFO’s recommendations for therapies in the combined “for consideration” and “recommended” categories.

For 124 cases of metastatic disease, however, Watson and its human counterparts were more frequently at odds, with only a 46% concordance.

In a subset analysis by receptor status, the investigators found that Watson was best – that is, most highly concordant – in triple-negative disease, with a 67.9% concordance in regard to MTB choices. In contrast, for patients with metastatic disease negative for the human epidermal growth factor receptor-2 (HER2), the concordance rate was a low 35%.

Dr. Somashekhar commented that part of the explanation for the discordant concordance rates by tumor subtype can be attributed to the fact that patients with triple-negative breast cancers have fewer treatment options than patients with HER2 negative–only tumors, who have a much broader array of possibilities.

The investigators also saw an overall concordance of WFO with the MTB recommendation at the time of the initial decision of 73%. But following 2016 MTB re-review of the original nonconcordant cases, the concordance between machine and mankind increased to 90%.

One area where WFO has its two-legged colleagues beaten hands down, however, is in the time it takes to capture and analyze data: humans took a mean of 20 minutes, Watson a median of 40 seconds.

The investigators acknowledged that WFO represents a further step toward personalized medicine, but emphasized that software will never replace the doctor-patient relationship or override the treating physician’s decisions.

‘This one would actually help’

At the briefing, moderator C. Kent Osborne, MD, from Baylor College of Medicine, Houston, commented, “I guess this not going to put us out of business as physicians – at least I hope not.”

Asked whether WFO could be a useful clinical tool or just another nuisance task added to an already overcrowded clinic schedule, Dr. Osborne responded, “I think this one would actually help.”

Anne Blaes, MD, from the University of Minnesota, Minneapolis, who also presented data at the briefing but was not involved in the concordance study, commented that, for oncologists practicing in the community, WFO offers the opportunity to get easy access to comprehensive, high-level information and expert opinions usually available only to those who toil in academic medical centers.

This study was investigative and received no external funding. Dr. Somashekhar, Dr. Osborne, and Dr. Blaes reported no conflicts of interest.