---
title: >-
  Estimating recurrent complication burden under time-varying risk: a comparison
  of Poisson regression and the partitioned count estimator
description: >-
  Yasmmyn Salinas · with Teresa Bufford, Derek Gazis, Julie Crawford, Kathleen
  Hurwitz, Alan Brookhart
date: '2026-09-02'
category: Events
canonical_url: https://www.pedestalhealth.com/resources/events/other/ispe-2026-poster-c-439/
source: Pedestal Health
license: © 2026 Pedestal Health. All rights reserved.
slug: ispe-2026-poster-c-439
id: 38hwhD9rRrvUOG7HMnnfkO
contentType: article
---

ISPE 2026 · Milan · Booth 300

# One dataset. Two estimators. Different long-term burden.

When complication risk changes over time, standard Poisson models can overstate long-term burden.


![ISPE 2026 event hero background](https://images.ctfassets.net/h4s3ip99qawo/1eizKn9Q3Fv3qgv7EmEdaM/a22337a6d457b68216faa2ece6690757/ispe-2026-event-background.png?w=1920&fm=webp&q=75)

## The estimator you choose changes the burden you project

**Key Numbers**

Anyone using registry data to project long-term complication risk, clinicians and HEOR teams alike, should care: the model you pick meaningfully changes the estimated burden, and that distorts downstream comparative-effectiveness and burden-of-disease analyses.

In the TARGET-IBD registry, we compared cumulative complications between 412 adults with Crohn's disease initiating anti-TNFα therapy and 610 initiating other advanced therapies (vedolizumab, ustekinumab, tofacitinib, natalizumab, ozanimod), using both Poisson regression and the partitioned count estimator with inverse probability of treatment weighting to adjust for confounding. By year 5, weighted Poisson estimated 68 complications per 100 anti-TNFα initiators versus 58 per 100 with the partitioned count estimator, because the event rate fell nearly 7-fold, from 27.5 per 100 person-years in year 1 to 4.08 by year 5. The pattern held in the other-AT arm as well (51 vs. 39 per 100 patients at year 5).

Poisson assumes a constant rate; that assumption did not hold here. The partitioned count estimator lets the rate vary, which is why the two methods disagree.



- **1,022 adults with Crohn's disease** initiating advanced therapy in the TARGET-IBD registry [anti-TNFα (n=412) vs. other AT (n=610)].
- **6.7-fold rate decline over follow-up:** 27.5 in year 1 vs. 4.08 events per 100 person-years by year 5.
- **Poisson overestimated 5-year burden in both arms:** anti-TNFα: 68 vs. 58 · other AT: 51 vs. 39 (per 100 patients, IPTW-adjusted; Poisson vs. partitioned count).

Salinas YD, Bufford T, Gazis D, Crawford J, Hurwitz KE, Brookhart MA. Estimating recurrent complication burden under time-varying risk: a comparison of Poisson regression and the partitioned count estimator. Poster C-439, 42nd ISPE Annual Meeting; August 29–September 2, 2026; Milan, Italy.

### C-439 · Session C

**Poster**

### August 29 – September 2, 2026

**When**

### Booth 300 · Allianz MiCo

**Where**

## Schedule a meeting

Tell us what you're working on and we'll hold a time to talk at Booth 300.



By submitting, you consent to Pedestal Health contacting you about this request and related work.

*Interactive embed: Pardot Form.*