Decision Fatigue Among Physicians

Transcription

1 Decision Fatigue Among Physicians Han Ye, Junjian Yi, Songfa Zhong 0 / 50

2 Questions Why Barack Obama in gray or blue suit? Why Mark Zuckerberg in gray T-shirt? 1 / 50

3 Questions Why Barack Obama in gray or blue suit? Why Mark Zuckerberg in gray T-shirt? Obama: You ll see I wear only gray or blue suits. I m trying to pare down decisions. I don t want to make decisions about what I m eating or wearing. Because I have too many other decisions to make. Zuckerberg: I really want to clear my life to make it so that I have to make as few decisions as possible about anything except how to best serve this community. 2 / 50

4 Research Questions Whether and how decision fatigue affects physician behavior? What kind of physicians are more vulnerable to decision fatigue? 3 / 50

5 Literature: Decision Fatigue Making too many decisions depletes individuals executive function and mental resources, which may influence their subsequent decisions (Baumeister et al., 1998; Stanovich and West, 2000; Kahneman, 2011; Baumeister and Tierney, 2012). Consumer purchasing for custom-made products (Levav et al., 2010) Judicial parole decisions (Danziger et al., 2011) Voter behavior (Augenblick et al., 2015) Financial analysts forecasts (Hirshleifer et al., 2017) 4 / 50

6 Literature: Physician Behavior Physicians decisions are affected by a variety of factors irrelevant to patient health, which may lead to large variations in procedure use, medical expenditure and patient outcomes. Peer effects among doctors and organizational culture (Lee and Mongan, 2009) Medical liability system (Currie and McLeod, 2008; Frakes, 2013) Physician beliefs about treatment (Cutler et al., 2013) Physicians financial incentives (Clemens and Gottlieb, 2014) Quality of physicians human capital (Currie and McLeod, 2017) 5 / 50

7 Patient Flow Source: The hospital A&E website 6 / 50

8 Patient Flow Data An entire set of ED visits from January 2011 to December 2012 in an acute general hospital of SG (264,115 patient cases). For each patient case, we observe detailed timestamps (e.g. the start and end times of triage, consultation, and task orders) and physician identifiers. With these data, we are able to reconstruct Real-time patient flow volume in the ED Patient s entire path through the ED Physician s shift schedules Sequences of actual patients who are seen by the physician in each shift Potential patients who arrive during a time when the physician s shift is in progress 7 / 50

9 Patient Information Basic demographics: birthdate, gender, address Patient acuity category scale (PACS) Arrival mode: ambulance or walk in Disposition type: discharge home, follow-up in primary care, inpatient admission... 8 / 50

10 Sample Restrictions All patient cases whose attending physician has at least 10 shifts observed is working in a shift with shift length 6-16 hours 242,761 patient cases, with 124 physician identities 9 / 50

11 Dependent Variables Physician decisions 1. physician discharge decision (dummies) outpatient disposition (discharge home or follow-up) discharge home inpatient admission 2. task orders total number of task orders (treatments and diagnostic tests) number of diagnostic tests 3. patient length of stay from the start to the end of patient case consultation Patient outcomes (dummies) 1. death in the ED 2. return visits to the ED within 14 days 10 / 50

12 Independent Variables: Decision Fatigue 1. Number of patients seen by the physician prior to the indexed patient s arrival 2. Hours relative to the shift beginning 11 / 50

13 Table: Summary statistics of outcome variables 12 / 50

14 Table: Summary statistics 13 / 50

15 Graphic Evidence A B Hospital Admission Outpatient Disposition Discharge Home C D Number of Tests Number of Tasks E >20 Length of Stay(mins) F >20 Figure: Physician decisions over the number of previous cases 14 / 50

16 Regression Specifications The baseline regression describing the association between fatigue and physician decisions is as follows: Y ijt = F atigue ijt α + X i β + T t γ + ν j + ɛ ijt (1) Y ijt - decisions for or outcome of patient i, treated by physician j with consultation start time t F atigue ijt - number of cases seen by physician j prior to patient i s arrival X i - patient characteristics including gender, age (in quadratic form), and PACS index T t - time fixed effects: hour of day, day of week and month-year interactions ν j - physician fixed effects standard errors clustered at the physician level 15 / 50

17 OLS Results Table: Fatigue on physician decisions 16 / 50

18 Identification Is the assignment of patients to physicians random? patient side physician side hospital administrators 17 / 50

19 Identification Is the assignment of patients to physicians random? patient side physician side hospital administrators IV: number of hospital ambulance arrivals an important determinant for the physician s workload plausibly exogenous to the underlying health of a given patient isolate the causal effect of workload on physician decision making 17 / 50

20 2SLS Results Table: Fatigue on physician decisions 18 / 50

21 Break within the Shift Mental resources might be replenished by a short rest which increases glucose and leads to a positive mood (Danziger et al., 2011). Restricted sample of visits: the physician is in a shift with an at-least-one-hour break (not in charge of any patient case) break up a shift into two distinct sessions (before and after break) around 10%, 23,733 visits 19 / 50

22 Graphic Evidence A B Hospital Admission Outpatient Disposition Discharged Home C D Number of Tests Number of Tasks E >= >=15 Length of Stay(mins) F >= >=15 Before Break After Break Figure: Physician decisions over the number of previous within-session cases 20 / 50

23 Specification 1 Y ijt = α 1 T otalcase ij + α 2 Session2 ij + α 3m D ijm m (2) +X i β + T t γ + ν j + ɛ ijt T otalcase ij - total number of cases seen by physician j prior to patient i s arrival Session2 - indicator for the after-break session D m - dummies indicating the first three cases in each session D 1-1st case in session 1 D 2-2nd case in session 1 D 3-3rd case in session 1 D 4-1st case in session 2 D 5-2nd case in session 2 D 6-3rd case in session 2 21 / 50

24 Table: Analysis using dummies for the first three decisions in a session 22 / 50

25 Specification 2 Y ijt = α 1 Case ij + α 2 Session2 ij + α 3 Case ij Session2 ij +X i β + T t γ + ν j + ɛ ijt (3) Case - number of previous cases within the current session Session2 - indicator for the after-break session Case Session2 - interaction term between Case and Session2 23 / 50

26 Table: Analysis of linear trend between sessions 24 / 50

27 Robustness 1: Alternative Fatigue Measure Alternative measure of fatigue - cumulative hours elapsed in the physician s shift Y ijt = m α m 1( t t(j, t) = m) + X i β + T t γ + ν j + ɛ ijt (4) t t(j, t): patient s arrival time relative to shift beginning α m : the average effect of m hours work (rounded up to the nearest nonnegative integer) on physician decisions reference category: visits arriving more than six hours after shift beginning 25 / 50

28 Table: Robustness 1 - Alternative fatigue measure 26 / 50

29 Table: Robustness 2 - Number of patients in the ED 27 / 50

30 Table: Robustness 3 - Include both the cumulative minutes and number of previous cases 28 / 50

31 Other Robustness Checks Restrictions on physician working hours - Shift length: 8-12 hours 8-10 hours 29 / 50

32 Heterogeneous Analysis 1 - Nonlinear Fatigue Effects Y ijt = m α m D ijm + X i β + T t γ + ν j + ɛ ijt (5) D ijm - dummy indicator: number of cases seen by physician j before patient i s arrival falls into group m D 1 : number of previous cases is between 0 to 2 D 2 : number of previous cases is between 3 to 5 D 3 : number of previous cases is between 6 to 8 D 4 : number of previous cases is between 9 to 11 D 5 : number of previous cases is between 12 to 14 D 6 : number of previous cases is between 15 to 17 D 7 : number of previous cases is between 18 to 20 reference category: number of previous cases is larger than / 50

33 Nonlinear Fatigue Effects Testcount lnlos Admitted Ordercount Outpatient Discharged #Previous cases Figure: Parameter estimates with 95% CI 31 / 50

34 Heterogeneous Analysis 2 Consider both quantity and composition of cases seen by the physician, before the indexed patient s arrival total number of cases: overall tendency number of severe cases: severe cases are likely to cost much more physician effort in terms of concentration and medical inputs 32 / 50

35 Table: Heterogeneous analysis 2 - number of previous severe cases 33 / 50

36 Will physician decision fatigue increase patient risk? 34 / 50

37 Physician Decision Fatigue and Patient Outcomes Percentage of Death A *** Percentage of Revisiting B *** Below mean Above mean Below mean Above mean Number of previous cases below (Left) versus above (Right) the sample mean (A) Data for PACS1&PACS2 cases. (B)Data for all cases. Figure: Patient outcomes over the number of previous cases 35 / 50

38 Table: Physician decision fatigue on patient outcomes 36 / 50

39 Who Are More Responsive to Fatigue? Physician specific fatigue effects obtained from IV estimations for each physician in our sample Physician characteristics gender, medical credentials, graduation year and school collected from Singapore Medical Council (SMC) 104 out of 124 physicians in our sample 37 / 50

40 Summary Statistics Table: Physician characteristics and fatigue effects 38 / 50

41 A B Discharged Home C >20 Graphs by physician experience Data for patients who were seen by physicians with medical experience (A) less than 10 years; (B) 10 to 17 years; (C) over 17 years. Figure: Proportion of discharge home over the number of previous cases 39 / 50

42 A B Number of Tests C >20 Graphs by physician experience Data for patients who were seen by physicians with medical experience (A) less than 10 years; (B) 10 to 17 years; (C) over 17 years. Figure: Average test orders over the number of previous cases 40 / 50

43 Table: Correlates of fatigue effects and physician characteristics 41 / 50

44 Conclusions Research questions Whether and how decision fatigue affects physician behavior? What kind of physicians are more vulnerable to decision fatigue? 42 / 50

45 Answers to Research Questions Decision fatigue: Physicians show an increased tendency to adopt outpatient disposition especially discharge home reduce task orders shorten patient length of stay increased ED mortality and return visits Physician characteristics influence the magnitudes of fatigue effects medical experience: U-shaped gender: Male 43 / 50

46 Implications To counteract the effects of decision fatigue more breaks; serious cases as earlier cases; more physicians and less choices. 44 / 50

47 Thank you for your comments! 44 / 50

48 Appendix Figures A Outpatient Disposition B C >20 Graphs by physician experience Data for patients who were seen by physicians with medical experience (A) less than 10 years; (B) 10 to 17 years; (C) over 17 years. Figure: Proportion of outpatient disposition over number of previous cases 45 / 50

49 A B Hospital Admission C >20 Graphs by physician experience Data for patients who were seen by physicians with medical experience (A) less than 10 years; (B) 10 to 17 years; (C) over 17 years. Figure: Proportion of hospital admission over the number of previous cases 46 / 50

50 A B Number of Tasks C >20 Graphs by physician experience Data for patients who were seen by physicians with medical experience (A) less than 10 years; (B) 10 to 17 years; (C) over 17 years. Figure: Average task orders over the number of previous cases 47 / 50

51 A Length of Stay(mins) B C >20 Graphs by physician experience Data for patients who were seen by physicians with medical experience (A) less than 10 years; (B) 10 to 17 years; (C) over 17 years. Figure: Average length of stay over the number of previous cases 48 / 50

52 Appendix Tables Table: Robustness - Restrictions on physician working hours shift length 8-12 hours 49 / 50

53 Table: Robustness - Restrictions on physician working hours shift length 8-10 hours 50 / 50