Comparison of Methods to Estimate Total Person-Time at Risk for Synthesis Studies of Incidence Rates in Observational Data
Author(s)
Clark L1, Zolotor A2, Curteis T1
1Costello Medical, Manchester, UK, 2Costello Medical, Boston, MA, USA
Presentation Documents
OBJECTIVES: Synthesis (e.g. meta-analysis) of incidence rates from observational data requires the total person-time at risk per study. While person-time can be calculated from individual participant data (IPD), for published aggregate data person-time is frequently unreported and requires estimation from the more commonly reported mean/median follow-up duration or Kaplan–Meier curves. There is no standard recommended method to estimate total person-time; here, we compare existing and novel methods.
METHODS: Brief targeted literature searches identified prostate cancer incidence studies fully reporting Kaplan–Meier curves, number at risk tables, median follow-up duration, and total person-time. Methods to estimate person-time were tested, including: multiplication of median follow-up by the number of patients; ’midpoint’ estimation, whereby the reported number of patients at risk in each interval is multiplied by the midpoint between the previous and succeeding timepoints, then summed; and calculation using pseudo-IPD from digitized Kaplan–Meier curves.
RESULTS: Three studies (Mehtälä [2020], Rompay [2019] and Seraphin [2021]) reporting the required data with different follow-up durations (14, 25 and 5 years, respectively) were identified and evaluated. Compared to reported total person-time, estimation from median follow-up yielded an under-estimate (range percentage difference vs reported: −43.23 to −24.07). Midpoint and pseudo-IPD estimation gave closer estimates (range percentage difference vs reported: −8.19 to 2.47 and −7.7 to 0.83, respectively).
CONCLUSIONS: When estimates of person-time are required (e.g. for incidence rate meta-analysis), midpoint and pseudo-IPD methods gave the best estimates. Estimation from median follow-up (the most frequent method in the literature) yielded substantial and consistent under-estimates of person-time, which inflates incidence rate estimates. Accordingly, midpoint or pseudo-IPD methods may be preferred as a unified method where data permits. Where data are limited to median follow-up, further research is required to determine whether estimation of person-time for synthesis of incidence is suitable, and potential biases should be acknowledged and explored via sensitivity analyses.
Conference/Value in Health Info
Value in Health, Volume 27, Issue 12, S2 (December 2024)
Code
MSR154
Topic
Epidemiology & Public Health, Methodological & Statistical Research, Study Approaches
Topic Subcategory
Literature Review & Synthesis, Meta-Analysis & Indirect Comparisons, Missing Data
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, Oncology