Diagnosing Physician Error: A Machine Learning Approach to Low-Value Health Care / Sendhil Mullainathan, Ziad Obermeyer.

By:

Mullainathan, Sendhil

Contributor(s):

Material type: Text

TextSeries: Working Paper Series (National Bureau of Economic Research) ; no. w26168.Publication details: Cambridge, Mass. National Bureau of Economic Research 2019.Description: 1 online resource: illustrations (black and white)Subject(s):

Online resources:

Available additional physical forms:

Hardcopy version available to institutional subscribers

Abstract: We use machine learning to study how physicians make decisions, in particular whether to test a patient for heart attack. We build an algorithmic model of patient risk, and identify cases where physician choices deviate from its predictions. To judge who was right, the algorithm or the physician, we use actual outcome data. Three findings emerge. First, many patients are tested who should not be: at typical cost-effectiveness thresholds, 62% of all tests would be cut. Second, many patients who should be tested are not. These patients go on to suffer adverse health events (including death) at rates exceeding clinical guidelines. A natural experiment using shift-to-shift testing variation confirms these findings: increased testing improves health and reduces mortality, but only for high-risk patients. We estimate the under-tested set to be 61% of the size of the tested population. Machine learning, by measuring value at the patient level rather than population level, is crucial for uncovering under- and over-testing. The simultaneous existence of both cannot easily be explained by physician incentives alone, and instead suggests physician errors. Third, we provide suggestive evidence on why physicians err: (i) they use too simple a model of risk, suggesting bounded rationality; (ii) they over-weight salient information; and (iii) they over-weight representative symptoms--those that fit the stereotype of heart attack. Together, these results suggest the need for health care models and policies to incorporate not just physician incentives, but also physician mistakes.

Tags from this library: No tags from this library for this title. Log in to add tags.

Average rating: 0.0 (0 votes)

Holdings
Item type	Home library	Collection	Call number	Status	Date due	Barcode	Item holds
Working Paper	Biblioteca Digital	Colección NBER	nber w26168 (Browse shelf(Opens below))	Not For Loan

Total holds: 0

Collection: Colección NBER Close shelf browser (Hides shelf browser)

Previous	No cover image available	No cover image available	No cover image available	No cover image available	No cover image available	No cover image available	No cover image available	Next
Previous	nber w26165 Predicting Consumer Default: A Deep Learning Approach /	nber w26166 Towards an Unstable Hook: The Evolution of Stock Market Integration Since 1913 /	nber w26167 Long-Term Macroeconomic Effects of Climate Change: A Cross-Country Analysis /	nber w26168 Diagnosing Physician Error: A Machine Learning Approach to Low-Value Health Care /	nber w26169 Mandatory CSR and Sustainability Reporting: Economic Analysis and Literature Review /	nber w2617 Flexible Staffing Arrangements and Employers' Short-Term Adjustment Strategies /	nber w26170 Does Condominium Development Lead to Gentrification? /	Next

August 2019.

We use machine learning to study how physicians make decisions, in particular whether to test a patient for heart attack. We build an algorithmic model of patient risk, and identify cases where physician choices deviate from its predictions. To judge who was right, the algorithm or the physician, we use actual outcome data. Three findings emerge. First, many patients are tested who should not be: at typical cost-effectiveness thresholds, 62% of all tests would be cut. Second, many patients who should be tested are not. These patients go on to suffer adverse health events (including death) at rates exceeding clinical guidelines. A natural experiment using shift-to-shift testing variation confirms these findings: increased testing improves health and reduces mortality, but only for high-risk patients. We estimate the under-tested set to be 61% of the size of the tested population. Machine learning, by measuring value at the patient level rather than population level, is crucial for uncovering under- and over-testing. The simultaneous existence of both cannot easily be explained by physician incentives alone, and instead suggests physician errors. Third, we provide suggestive evidence on why physicians err: (i) they use too simple a model of risk, suggesting bounded rationality; (ii) they over-weight salient information; and (iii) they over-weight representative symptoms--those that fit the stereotype of heart attack. Together, these results suggest the need for health care models and policies to incorporate not just physician incentives, but also physician mistakes.