Eleanor Pullenayegum

Background: Longitudinal data can be used to study disease progression and are often collected at irregular intervals. When the assessment times are informative about the severity of the disease, regression analyses of the outcome trajectory over time based on Generalized Estimating Equations (GEEs) result in biased estimates of regression coefficients. Inverse-intensity weighted GEEs (IIW-GEEs) are a popular approach to account for informative assessment times and yield unbiased estimates of outcome model coefficients when the assessment times and outcomes are conditionally independent given previously observed data. However, a consequence of irregular assessment times is that some patients may have no follow-up assessments at all, and it is common practice to omit these patients from analyses when studying the outcome trajectory over time. Methods: We show mathematically that IIW-GEEs yield biased estimates of regression coefficients when patients with no follow-up assessments are excluded from analyses. We design a simulation study to evaluate how the bias varies with sample size, assessment frequency, follow-up time, and the informativeness of the assessment time process. Using the STAR*D trial of treatments for major depressive disorder, we examine the extent of bias in practice. Results: Our simulation results showed the bias incurred by omitting patients with no follow-up visits increased as visit frequency decreased and as the duration of follow-up decreased. In the STAR*D trial, omitting patients with no follow-up visits led to over-estimation of the rate of improvement in depressive symptoms. Conclusions: Studies should be designed to ensure patients with no follow-up are included in the data. This can be achieved by a) creating inception cohorts; b) when taking sub-samples of existing cohorts, ensuring that patients without follow-up assessments are included; c) dropping exclusion criteria based on availability of follow-up visits.

Background: EQ-5D-3L health utilities inform health technology assessments in oncology, often derived from clinical trials enrolling highly selected populations, which may over-estimate real-world health utilities. Little is known about the association between socioeconomic status (SES) and EQ-5D-3L health utilities. Our objective was to examine EQ-5D-3L health utilities across SES in a real-world sample of patients with cancer. Methods: We conducted a cross-sectional analysis of EQ-5D-3L responses from 170 adult patients with cancer accrued from an Ontario oncology centre. We fitted multivariable linear regression models to estimate associations between covariates and EQ-5D-3L health utilities. Results: Lower EQ-5D-3L health utilities were significantly associated with the lowest and undisclosed family income categories in models with and without birth sex (p < 0.05). In the model including birth sex, disutility estimates for lowest family income (<CAD 29K) and undisclosed income was −0.202, 95% CI (−0.371 to −0.033), and −0.123 (−0.235 to −0.012), respectively. For the model excluding birth sex, disutility estimates for lowest income and undisclosed income was −0.163 (−0.280 to −0.046) and −0.106 (−0.184 to −0.028). Conclusions: Lower EQ-5D-3L health utilities were significantly associated with low income and nondisclosure, highlighting the need to incorporate SES in analyzing health utilities in cancer care.

Longitudinal data are often subject to irregular visiting times, with outcomes and visit times influenced by a latent variable. Semi‐parametric joint models that account for this dependence have been proposed; among these, the Sun model is the most suitable for count data as it employs a multiplicative link function. Semi‐parametric joint models define an intercept function as the mean outcome when all covariates are set to zero; this is differenced out in the course of estimation and is consequently not estimated. The Sun estimator thus provides estimates of relative covariate effects, but is unable to provide estimates of absolute effects or of longitudinal prognosis in the absence of covariates. We extend the Sun model by additionally estimating the intercept term, showing that our extended estimator is consistent and asymptotically Normal. In simulations, our estimator outperforms the original Sun estimator in terms of bias and standard error and is also more computationally efficient. We apply our estimator to a longitudinal study of tumor recurrence among bladder cancer patients. Provided the intercept term can be adequately captured using splines, we recommend that our extended Sun estimator be used in place of the original estimator, since it leads to smaller bias, smaller standard errors, and allows estimation of the mean outcome trajectories.

Cancer treatments impact health-related quality of life (HRQoL). EQ-5D is a standardized generic measure of HRQoL. The objective of this project was to assess, from the patient’s perspective, the feasibility of implementing real-world EQ-5D-3L measurement at a pilot site, as a first step to large-scale collection of EQ-5D from patients with cancer across Ontario. This was a prospective longitudinal study at a single oncology centre to understand the feasibility of routinely collecting EQ-5D-3L while patients receive chemotherapy (N = 170). Consenting patients completed an additional questionnaire on feasibility, and a subset of participants were directly interviewed to provide further feedback and suggest improvements to questionnaire collection. Themes emerging from the interviews were analyzed using content analysis. Of 170 eligible and consenting patients who completed an initial EQ-5D-3L questionnaire, 103 (60.6%) completed at least one follow-up questionnaire. When asked about willingness to answer future questionnaires at subsequent visits, 115 (67.3%) answered definitely; 35 (20.5%) very likely. Patients provided feedback on their overall experience of completing EQ-5D-3L, the questionnaire presentation, frequency of completion, and analytic plans. Patients expressed that routinely collecting EQ-5D-3L is feasible. Incorporating patient feedback regarding EQ-5D collection will facilitate implementation of systematic collection at cancer centres across Ontario.

Electronic health records (EHRs) provide an efficient approach to generating rich longitudinal datasets. However, since patients visit as needed, the assessment times are typically irregular and may be related to the patient's health. Failing to account for this informative assessment process could result in biased estimates of the disease course. In this paper, we show how estimation of the disease trajectory can be enhanced by leveraging an underutilized piece of information that is often in the patient's EHR: physician‐recommended intervals between visits. Specifically, we demonstrate how recommended intervals can be used in characterizing the assessment process and in investigating the sensitivity of the results to assessment not at random (ANAR). We illustrate our proposed approach in a clinic‐based cohort study of juvenile dermatomyositis (JDM). In this study, we found that the recommended intervals explained 78% of the variability in the assessment times. Under a specific case of ANAR where we assumed that a worsening in disease led to patients visiting earlier than recommended, the estimated population average disease activity trajectory was shifted downward relative to the trajectory assuming assessment at random. These results demonstrate the crucial role recommended intervals play in improving the rigor of the analysis by allowing us to assess both the plausibility of the AAR assumption and the sensitivity of the results to departures from this assumption. Thus, we advise that studies using irregular longitudinal data should extract recommended visit intervals and follow our procedure for incorporating them into analyses.

Objectives: Discrete choice experiments (DCEs) as a valuation method require preferences to be anchored on the quality-adjusted life-year scale, usually through tasks involving choices between immediate death and various impaired health states or between health states with varying durations of life. We sought to determine which anchoring approach aligns best with the composite time tradeoff (cTTO) method, with a view to informing a valuation protocol that uses DCEs in place of the cTTO. Methods: A total of 970 respondents from Trinidad and Tobago completed a DCE with duration survey. Tasks involved choosing between 2 lives with identical durations, followed by a third option, representing either full health for a number of years or immediate death. Data were analyzed using mixed logit models, both with and without exponential discounting for time preferences. Results: Assuming linear time preferences, the estimated utility of immediate death was −2.1 (95% credible interval [CrI] −3.2 to −1.2) versus −0.28 (95% CrI −0.47, −0.10) when allowing for nonlinear time preferences. Under linear time preferences, the predicted health-state values anchored on duration had range (−1.03, 1) versus (0.34, 1) when anchored on immediate death. The ranges under nonlinear time preferences were (−0.54, 1) versus (−0.22, 1). The estimated discount parameter was 23% (95% CrI 22% to 25%). Conclusions: The nonzero discount parameter indicates that time preferences were nonlinear. Nonlinear time preferences anchored on duration provided the closest match to the benchmark EQ-VT cTTO values in Trinidad and Tobago, whose range was (−0.6, 1). Thus, DCE with duration can provide similar values to cTTO provided that nonlinear time preferences are accounted for and anchoring is based on duration. Highlights: Time preferences for health states in Trinidad and Tobago were nonlinear. In discrete choice tasks, we show that immediate death has a utility less than zero. DCE utilities under nonlinear time preferences with anchoring on duration agreed well with cTTO utilities.

Bayesian methods are becoming increasingly in demand in clinical and public health comparative effectiveness research. Limited literature has explored parametric Bayesian causal approaches to handle time-dependent treatment and time-dependent covariates. In this article, building on to the work on Bayesian g-computation, we propose a fully Bayesian causal approach, implemented using latent confounder classes which represent the patient’s disease and health status. Our setting is suitable when the latent class represents a true disease state that the physician is able to infer without misclassification based on manifest variables. We consider a causal effect that is confounded by the visit-specific latent class in a longitudinal setting and formulate the joint likelihood of the treatment, outcome and latent class models conditionally on the class indicators. The proposed causal structure with latent classes features dimension reduction of time-dependent confounders. We examine the performance of the proposed method using simulation studies and compare the proposed method to other causal methods for longitudinal data with time-dependent treatment and time-dependent confounding. Our approach is illustrated through a study of the effectiveness of intravenous immunoglobulin in treating newly diagnosed juvenile dermatomyositis.

The use of EQ-5D instruments in clinical, policy and economic applications continues to grow internationally. Population norms studies provide baseline values against which demographic and patient groups are compared and inequality is assessed. This study presents updated EQ-5D-5L population norms for 2022–2023, evaluates inequality and compares the results with those of 2012. Demographic and EQ-5D-5L data were obtained from mutually exclusive, representative samples of adults in three studies conducted from July 2022 through May 2023. EQ-5D-5L index values, EQ VAS scores, and ceilings (all dimensions at level 1) were calculated for age-sex groups and stratifiers including education, income, ethnicity, marital status, and employment status. For inequality, the Kakwani index was calculated for the EQ VAS scores and index values, and ordered logit models were used to obtain odds ratios for reporting higher levels of problems on each dimension for demographic groups. The results were compared with those from 2012 which included applying the value set that had been used for the 2022–2023 population norms to the 2012 states. Data were obtained form 2,989 respondents. The mean index value was 0.921, EQ VAS was 79.6 and the ceiling was 31.5%. The dimensions with the highest rates of reported problems at any level (2–5) were pain/discomfort (43%) and anxiety/depression (39%). The Kakwani index was 0.113 for EQ VAS and 0.058 for index values, with sex accounting for the largest relative contribution. Mean index values, EQ VAS scores, and ceilings were lower across all demographic groups in 2022–2023 compared to 2012. This is the first study to investigate how EQ-5D-5L population norms have changed within a country over time. Significant changes were observed in the EQ-5D-5L measures and the relative frequencies of reported problems on the dimensions. Inequality increased, and there were changes in the levels of reported problems on the dimensions for demographic groups. Such changes suggest that national population norms should be updated periodically to capture changes in health status, perceptions of health, and health inequality.