Background
Supportive and palliative care
Mortality prediction
Objective
Methods
Data
Study setting
Patient population
Mortality outcomes
Feature construction
-
Count of each unique code,
-
Count of unique codes and total code count across days, and
-
Mean, variance, minimum, maximum, and range of the daily number of codes.
Experimental design
Model development
Implementation
Operating threshold
Prospective validation
Evaluation in the context of potential demographic Bias
Results
Patient cohort and outcomes
All Patients n = 128,941 | Training Set n = 72,437 | Testing Set n = 46,458 | |||
---|---|---|---|---|---|
Demographics a | |||||
Measure | Value | ||||
Age | % (n) | % (n) | % (n) | * | |
18–29 | 11.5% (14786) | 10.7% (7778) | 13.1% (6087) | ||
30–39 | 17.5% (22607) | 18.0% (13053) | 18.0% (8361) | ||
40–49 | 9.45% (12183) | 9.49% (6877) | 9.69% (4504) | ||
50–59 | 13.3% (17204) | 13.5% (9784) | 13.4% (6206) | ||
60–69 | 18.2% (23500) | 18.7% (13556) | 17.3% (8026) | ||
70–79 | 15.8% (20388) | 15.8% (11439) | 15.1% (7008) | ||
80–89 | 10.7% (13839) | 10.5% (7588) | 10.2% (4748) | ||
90+ | 3.44% (4434) | 3.26% (2362) | 3.27% (1518) | ||
Ethnicity b | % (n) | % (n) | % (n) | * | |
Hispanic | 9.75% (3467) | 9.77% (2336) | 8.62% (666) | ||
Not Hispanic | 90.3% (32086) | 90.2% (21584) | 91.4% (7060) | ||
Unknown | -- (93388) | -- (48517) | -- (38732) | ||
Race | % (n) | % (n) | % (n) | * | |
Black | 10.9% (14033) | 11.0% (7933) | 10.7% (4987) | ||
East Asian | 7.38% (9520) | 6.50% (4707) | 9.10% (4230) | ||
West Asian | 1.66% (2146) | 1.68% (1219) | 1.74% (807) | ||
White | 61.6% (79424) | 64.1% (46404) | 57.3% (26642) | ||
Other | 16.4% (21181) | 14.8% (10692) | 18.8% (8714) | ||
Unknown | 2.05% (2637) | 2.05% (1482) | 2.32% (1078) | ||
Sex | % (n) | % (n) | % (n) | ||
Female | 60.1% (77478) | 60.3% (43664) | 60.5% (28130) | ||
Male | 39.9% (51459) | 39.7% (28770) | 39.4% (18327) | ||
Unknown | 0% (4) | 0% (3) | 0% (1) | ||
Site | % (n) | % (n) | % (n) | * | |
Tisch | 63.4% (81807) | 72.3% (52398) | 49.2% (22877) | ||
Orthopedic | 15.6% (20137) | 18.1% (13122) | 12.8% (5938) | ||
Brooklyn | 20.9% (26997) | 9.55% (6917) | 38% (17643) | ||
Outcomes c | % (n) | % (n) | % (n) | ||
Any known death | 7.93% (10229) | 9.00% (6521) | 5.20% (2414) | * | |
60-day death | 4.15% (5356) | 4.05% (2935) | 3.57% (1657) | * | |
Median [IQR] | Median [IQR] | Median [IQR] | |||
Days from admission to death | 53 [6, 205] | 83 [12, 306] | 21 [1, 92.75] | * | |
Comorbidities d | Median [IQR] | Median [IQR] | Median [IQR] | ||
Charlson Score | 1 [0, 2] | 1 [0, 2] | 0 [0, 2] | * | |
% (n) | % (n) | % (n) | |||
AIDS/HIV | 0.626% (635) | 0.61% (349) | 0.506% (176) | ||
Cancer (any malignancy) | 16.8% (17094) | 18.2% (10432) | 13.2% (4594) | * | |
Cerebrovascular disease | 10.0% (10149) | 9.99% (5716) | 8.13% (2826) | * | |
Chronic obstructive pulmonary disease | 17.9% (18218) | 18.6% (10649) | 13.5% (4703) | * | |
Congestive heart failure | 12.0% (12144) | 11.8% (6774) | 8.56% (2978) | * | |
Dementia | 3.67% (3721) | 3.18% (1819) | 3.09% (1075) | ||
Diabetes with chronic complications | 6.34% (6439) | 4.9% (2806) | 5.68% (1977) | * | |
Diabetes without chronic complications | 16.8% (17019) | 16.2% (9256) | 14.4% (4995) | * | |
Hemiplegia or paraplegia | 2.92% (2962) | 2.83% (1617) | 2.35% (817) | * | |
Metastatic solid tumor | 6.02% (6115) | 6.39% (3657) | 4.55% (1584) | * | |
Mild liver disease | 6.40% (6495) | 6.23% (3566) | 5.14% (1787) | * | |
Moderate or severe liver disease | 1.62% (1642) | 1.59% (910) | 1.11% (385) | * | |
Myocardial infarction | 9.73% (9874) | 9.48% (5423) | 6.9% (2400) | * | |
Peptic ulcer disease | 1.84% (1871) | 1.76% (1009) | 1.27% (443) | * | |
Peripheral vascular disease | 13.1% (13278) | 13.0% (7446) | 9.97% (3469) | * | |
Renal disease | 10.9% (11093) | 10.4% (5937) | 7.93% (2759) | * | |
Rheumatoid disease | 2.87% (2915) | 3.11% (1781) | 2.06% (718) | * | |
Predictors | |||||
Range | Measure | Median [IQR] | Median [IQR] | ||
1–30 days | # of diagnoses | 3 [0, 12] | 3 [0, 13] | 2 [0, 10] | * |
1–30 days | # of lab results | 0 [0, 46] | 3 [0, 47] | 0 [0, 43] | * |
1–30 days | # of office visits | 3 [1, 6] | 3 [1, 6] | 2 [1, 5] | * |
1–30 days | # of emergency department visits | 0 [0, 0] | 0 [0, 0] | 0 [0, 0] | * |
1–30 days | # of hospitalizations | 0 [0, 0] | 0 [0, 0] | 0 [0, 0] | * |
1–365 days | # of diagnoses | 15 [2, 51] | 14 [2, 52] | 11 [0, 36] | * |
1–365 days | # of lab results | 35 [0, 151] | 34 [0, 142] | 15 [0, 84] | * |
1–365 days | # of office visits | 11 [5, 25] | 11 [5, 25] | 9 [4, 20] | * |
1–365 days | # of emergency department visits | 0 [0, 1] | 0 [0, 1] | 0 [0, 1] | |
1–365 days | # of hospitalizations | 0 [0, 1] | 0 [0, 1] | 0 [0, 0] | * |
Retrospective modeling
Performance within the training cohort
Model | Cohort | Measure | AUROC | AUPRC |
---|---|---|---|---|
Lasso regression | Training (Cross-validation) | Mean [min, max] | 78.8 [78.0, 80.2] | 21.0 [18.3, 22.0] |
XGBoost | Training (Cross-validation) | Mean [min, max] | 84.6 [83.8, 86.0] | 25.7 [21.2, 27.4] |
Random forest | Training (Cross-validation) | Mean [min, max] | 86.9 [85.3, 87.7] | 26.4 [20.1, 31.0] |
Testing (Bootstrapped) | Median [95% CI] | 87.2 [86.1, 88.2] | 28.0 [25.0, 31.0] | |
Brooklyn | Median [95% CI] | 83.8 [81.9, 85.6] | 26.6 [22.5, 31.0] | |
Non-Brooklyn | Median [95% CI] | 88.9 [87.5, 90.2] | 30.1 [26.4, 33.7] |