Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Time-to-Event Estimation with Unreliably Reported Events in Medicare Health Plan Payment

Oana M. Enache, Sherri Rose

TL;DR

This work tackles time-to-event estimation when event reporting is unreliable in Medicare Advantage, focusing on incident coding of hierarchical condition categories (HCCs) under CMS-HCC risk adjustment. It extends restricted mean time lost (RMTL) methods to accommodate competing risks and uses incident coding as the event of interest across MA and traditional Medicare (TM), proposing estimators for underreporting and possible severity-based upcoding, along with a publicly available R package to simulate labeled data. Simulation studies demonstrate that the proposed estimators recover differences in time-to-incident coding between MA and TM and are less sensitive to underreporting than a DECI-like comparator, highlighting policy-relevant potential to monitor coding practices over time. The open-source tools and realistic simulations enable scalable evaluation of risk-adjustment incentives and could inform upgrades to reduce unnecessary Medicare spending while preserving beneficiary value.

Abstract

Time-to-event estimation (i.e., survival analysis) is common in health research, most often using methods that assume proportional hazards and no competing risks. Because both assumptions are frequently invalid, estimators more aligned with real-world settings have been proposed. An effect can be estimated as the difference in areas below the cumulative incidence functions of two groups up to a pre-specified time point. This approach, restricted mean time lost (RMTL), can be used in settings with competing risks as well. We extend RMTL estimation for use in an understudied health policy application in Medicare. Medicare currently supports healthcare payment for over 69 million beneficiaries, most of whom are enrolled in Medicare Advantage plans and receive insurance from private insurers. These insurers are prospectively paid by the federal government for each of their beneficiaries' anticipated health needs using an ordinary least squares linear regression algorithm. As all coefficients are positive and predictor variables are largely insurer-submitted health conditions, insurers are incentivized to upcode, or report more diagnoses than may be accurate. Such gaming is projected to cost the federal government $40 billion in 2025 alone without clear benefit to beneficiaries. We propose several novel estimators of coding intensity and possible upcoding in Medicare Advantage, including accounting for unreliable reporting. We demonstrate estimator performance in simulated data leveraging the National Institutes of Health's All of Us study and also develop an open source R package to simulate realistic labeled upcoding data, which were not previously available.

Time-to-Event Estimation with Unreliably Reported Events in Medicare Health Plan Payment

TL;DR

This work tackles time-to-event estimation when event reporting is unreliable in Medicare Advantage, focusing on incident coding of hierarchical condition categories (HCCs) under CMS-HCC risk adjustment. It extends restricted mean time lost (RMTL) methods to accommodate competing risks and uses incident coding as the event of interest across MA and traditional Medicare (TM), proposing estimators for underreporting and possible severity-based upcoding, along with a publicly available R package to simulate labeled data. Simulation studies demonstrate that the proposed estimators recover differences in time-to-incident coding between MA and TM and are less sensitive to underreporting than a DECI-like comparator, highlighting policy-relevant potential to monitor coding practices over time. The open-source tools and realistic simulations enable scalable evaluation of risk-adjustment incentives and could inform upgrades to reduce unnecessary Medicare spending while preserving beneficiary value.

Abstract

Time-to-event estimation (i.e., survival analysis) is common in health research, most often using methods that assume proportional hazards and no competing risks. Because both assumptions are frequently invalid, estimators more aligned with real-world settings have been proposed. An effect can be estimated as the difference in areas below the cumulative incidence functions of two groups up to a pre-specified time point. This approach, restricted mean time lost (RMTL), can be used in settings with competing risks as well. We extend RMTL estimation for use in an understudied health policy application in Medicare. Medicare currently supports healthcare payment for over 69 million beneficiaries, most of whom are enrolled in Medicare Advantage plans and receive insurance from private insurers. These insurers are prospectively paid by the federal government for each of their beneficiaries' anticipated health needs using an ordinary least squares linear regression algorithm. As all coefficients are positive and predictor variables are largely insurer-submitted health conditions, insurers are incentivized to upcode, or report more diagnoses than may be accurate. Such gaming is projected to cost the federal government $40 billion in 2025 alone without clear benefit to beneficiaries. We propose several novel estimators of coding intensity and possible upcoding in Medicare Advantage, including accounting for unreliable reporting. We demonstrate estimator performance in simulated data leveraging the National Institutes of Health's All of Us study and also develop an open source R package to simulate realistic labeled upcoding data, which were not previously available.
Paper Structure (31 sections, 2 equations, 3 figures)

This paper contains 31 sections, 2 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Restricted mean survival time-based methods
  3. Payment formulas in Medicare Advantage
  4. Contributions
  5. Methods
  6. Notation
  7. Incident events
  8. Reference events
  9. Target estimands
  10. Underreporting in comparison group
  11. Difference in mean time without event across groups
  12. Possible severity-based upcoding
  13. Estimators
  14. Underreporting in comparison group
  15. Difference in mean time without event across groups
  16. ...and 16 more sections

Figures (3)

  • Figure 1: Cumulative incidence functions for the Specified Heart Arrhythmias Hierarchical Condition Category (HCC) in simulated Medicare Advantage (MA) and Traditional Medicare (TM) groups: 20% any-available MA upcoding. Specified heart arrhythmias corresponds to HCC238, which does not have any competing events. For this HCC, 20% of any-available individuals in the MA group are upcoded and 5% of any-available individuals in the TM comparison group are upcoded. The first monitoring period is labeled M1, and the second monitoring period is labeled M2. Given the large sample size, confidence intervals are very narrow and are therefore omitted as they cannot be distinguished visually.
  • Figure 2: Within-monitoring period period$\boldsymbol{\psi}$estimatesfor the Specified Heart Arrhythmias Hierarchical Condition Category (HCC) in simulated Medicare Advantage (MA) and Traditional Medicare (TM) groups. Specified heart arrhythmias corresponds to HCC238, which does not have any competing events. For this HCC, any-available individuals in the MA group are upcoded to varying degrees and any-available individuals in the TM comparison group are upcoded 5%. Given the large sample size, confidence intervals are very narrow and are therefore omitted as they cannot be distinguished visually.
  • Figure 3: $\textbf{DECI}^\dagger$estimate across all Medicare Advantage (MA) Version 28 Hierarchical Condition Categories (HCCs) at varying degrees of upcoding and underreporting in simulated MA and Traditional Medicare (TM) groups. Three separate degrees of MA group upcoding occur sequentially over each monitoring period in HCC238 (any-available) and HCC125 (lower severity) only. The TM group is upcoded 5% sequentially for the same HCCs over equivalent periods. Given the large sample size, confidence intervals are very narrow and are therefore omitted as they cannot be distinguished visually.