Impact of Smoothing and Modeling Approach on Quality-Adjusted Life Expectancy Estimates

Quality-adjusted life expectancy (QALE) is a composite indicator integrating life expectancy and health utility values. Most studies have used the Sullivan method to calculate QALE, whereas Markov modeling offers a more flexible alternative simulating health transitions over time. The primary objective of this study was to estimate age- and sex-specific QALE for Quebec and to compare results across 4 methodological approaches: Sullivan versus Markov modeling, each with and without cubic polynomial fit of age-specific utilities.

We analyzed 4803 EQ-5D-5L records from 2016 to 2024 health surveys and pooled 2021 to 2023 life tables. Age-specific utilities were smoothed using cubic polynomial regression. Age- and sex-specific QALE norms were estimated using both Sullivan method and a stochastic 2-state Markov microsimulation, with Monte Carlo simulations applied to both methods to quantify uncertainty. Sensitivity analyses assessed the impact of reducing the Markov cycle length from 1 year to 0.5 year for the combined population, and differences between methods were evaluated using a 1-sample t test.

Cubic polynomial regressions produced smooth age-utility curves with excellent fit for the combined population (R² = 0.86), males (R² = 0.96), and females (R² = 0.92). Smoothed utilities modestly reduced uncertainty, particularly at older ages. Total QALE estimates from the Sullivan and Markov approaches were highly consistent, with small absolute differences (0.28-0.32 QALE) and strong correlation (R² = 0.99) over the full remaining lifetime. The 1-sample t test showed that Markov QALE estimates were slightly higher than Sullivan estimates (mean difference 0.315 QALE, 95% CI 0.308-0.322; t = 85.47, P < .001), though the absolute differences were minor relative to overall QALE. Sensitivity analyses demonstrated that reducing the Markov cycle length had minimal impact on QALE estimates (differences ≤0.29 QALE), confirming robustness.

These findings support the use of either method for population health assessment and health technology evaluation because both produce valid and reliable QALE estimates.