Learning through Capacity Awareness to Minimize Hospital Readmissions
Organizations: Princeton University · Purdue University · University of Southern California
Abstract
Predictive algorithms are increasingly used to guide hospital discharge decisions, yet these models can exhibit censoring bias for vulnerable patients due to insufficient historical data or past discharge decisions. When biased readmission risk predictors are used to inform discharges, the training data generated by discharge decisions may perpetuate similar biases when predictors are re-trained, leading to a vicious cycle. We extend a queueing-based Markov decision process for hospital discharge to incorporate periodically retrained readmission risk predictors. Our key insight is that cost-optimal, "capacity-aware" operational decision policies can naturally diversify the lengths of stay at which patient outcomes are observed, enabling consistent estimation of readmission risk without explicit fairness constraints. We prove that repeated retraining under discharge-up-to policies converges to unbiased risk estimates under well-specification. In synthetic experiments and calibrated simulations based on patient-level data from a partner hospital, we find that where commonly used threshold-on-risk policies can remain persistently biased, discharge-up-to policies reduce learning error over status quo decision policies (such as thresholding on risk) by 24%-25%. These improvements to estimation also mitigate some forms of bias, reducing readmission cost disparities (adjusted for baseline risk differences between demographic groups) by 38%-63%. Our results demonstrate how capacity-aware decision policies can improve both predictive performance and fairness by addressing censoring bias.
Figures & tables
| Method | rMSE | Cost | ||
|---|---|---|---|---|
| Dis* ( ) | 0.095 | 0.0113 | 558.9974 | 0.0538 |
| Thresh | 0.1212 ( ) | 0.0110 ( ) | 854.8178 | 0.0862 ( ) |
| Improvement over Thresh | 24.4% ( ) | -2.49% ( ) | 34.4% ( ) | 40.3% ( ) |
| State ( ) | 0.123 | 0.0114 | 1212.6527 | 0.1439 |
| Improvement over State | 25.5% ( ) | 0.8% ( ) | 37.7% ( ) | 62.7% ( ) |
Appendix figures & tables12 assets
Supplementary material from the paper’s appendix.
Appendix
| Notation | Description |
|---|---|
| , | Entrywise norm |
| Number of arrivals on day for patients of type | |
| Number of arrivals of all types, | |
| Expected readmission cost of discharges adjusted for baseline risk (risk if held until maximum LOS), | |
| Expected readmission cost of discharges from group adjusted for baseline risk | |
| Disparity in adjusted expected readmission cost between and for round . See Appendix C.1 for definition. |
| Policy Name | Mean Total Cost ( ) | SEM Total Cost | Mean rMSE | SEM rMSE |
|---|---|---|---|---|
| Dis ( ) | 180.3612 | 0.0117 | 0.0002 | |
| Dis ( ) | 170.6799 | 0.0082 | 0.0002 | |
| Dis ( ) | 177.2092 | 0.0067 | 0.0001 | |
| Dis ( ) | 188.2904 | 0.0062 | 0.0001 | |
| Dis ( ) | 209.0363 | 0.0072 | 0.0001 | |
| Dis ( ) | 327.2783 | 0.0098 | 0.0002 |
| Policy Name | Mean Cost ( ) | SEM Total Cost | Mean rMSE | SE |
|---|---|---|---|---|
| Dis ( ) | 588.5656 | 0.0118 | 0.0015 | |
| Dis ( ) | 435.4185 | 0.0093 | 0.001 | |
| Dis ( ) | 652.0527 | 0.007 | 0.0006 | |
| Dis ( ) | 819.1144 | 0.0067 | 0.0009 | |
| Dis ( ) | 1006.7915 | 0.0076 | 0.0008 | |
| Dis ( ) | 2295.5058 | 0.0108 | 0.0014 |
| Policy Name | Mean | SEM | Mean | SEM |
|---|---|---|---|---|
| Dis ( ) | 0.0982 | 0.0008 | -0.0037 | 0.0016 |
| Dis ( ) | 0.0845 | 0.0009 | -0.0094 | 0.0016 |
| Dis ( ) | 0.0769 | 0.001 | -0.0059 | 0.0012 |
| Dis* ( , ) | 0.0845 | 0.0009 | -0.0043 | 0.0014 |
| Dis* ( , ) | 0.0978 | 0.0006 | -0.01 | 0.0016 |
| Dis* ( , ) | 0.0817 | 0.0009 | -0.0069 | 0.0013 |
| Policy Name | Mean Cost ( ) | SEM Cost ( ) | Mean Cost ( ) | SEM Cost ( ) |
|---|---|---|---|---|
| Dis ( ) | 541.6881 | 6.0977 | 146.7497 | |
| Dis ( ) | 832.7707 | 7.7814 | 131.5977 | |
| Dis ( ) | 1344.1148 | 8.3699 | 138.142 | |
| Dis* ( , ) | NA | NA | 165.1165 | |
| Dis* ( , ) | 555.0492 | 6.1148 | NA | NA |
| Dis* ( , ) | NA | NA | 183.4523 |
| Policy Name | Mean ( ) | SEM ( ) | Mean ( ) | SEM ( ) |
|---|---|---|---|---|
| Dis ( ) | -0.1976 | 0.0332 | -4.9409 | 0.8305 |
| Dis ( ) | -0.2584 | 0.0353 | -6.4595 | 0.8824 |
| Dis ( ) | -0.1809 | 0.0325 | -4.5226 | 0.8119 |
| Dis* ( , ) | NA | NA | -3.8986 | 0.852 |
| Dis* ( , ) | -0.2309 | 0.0443 | NA | NA |
| Dis* ( , ) | NA | NA | -4.4796 | 1.033 |
| Policy Name | Mean ( ) | SEM ( ) | Mean ( ) | SEM ( ) |
|---|---|---|---|---|
| Dis ( ) | 0.0642 | 0.0078 | 1.6053 | 0.196 |
| Dis ( ) | 0.1278 | 0.0093 | 3.1949 | 0.2328 |
| Dis ( ) | 0.1513 | 0.01 | 3.7824 | 0.2498 |
| Dis* ( , ) | NA | NA | 2.7694 | 0.3119 |
| Dis* ( , ) | 0.0409 | 0.0086 | NA | NA |
| Dis* ( , ) | NA | NA | 2.893 | 0.2753 |
| Policy 1 | Policy 2 | Metric | % | Decrease/Increase | -value |
|---|---|---|---|---|---|
| Dis* ( , ) | Thresh ( ) | 33% | Dec. | ||
| Dis* ( , ) | State ( ) | 34% | Dec. | ||
| Dis* ( , ) | Thresh ( ) | 11% | Dec. | 0.53 | |
| Dis* ( , ) | State ( ) | 23% | Dec. | 0.47 | |
| Dis* ( , ) | Thresh ( ) | Cost | 35% | Dec. | |
| Dis* ( , ) | State ( ) | Cost | 54% | Dec. |
Healthcare
Healthcare AI develops and evaluates models and systems for clinical care, medical research, public health, and healthcare operations.
See all 640 papers