A systematic review of the quality of reporting of simulation studies about methods for the analysis of complex longitudinal patient-reported outcomes data

A systematic review of simulation studies which compare existing statistical methods to account for non-compliance in randomised controlled trials

INTRODUCTION

Non-compliance is a common challenge for researchers and may reduce the power of an intention-to-treat analysis. Whilst a per protocol approach attempts to deal with this issue, it can result in biased estimates. Several methods to resolve this issue have been identified in previous reviews, but there is limited evidence supporting their use. This review aimed to identify simulation studies which compare such methods, assess the extent to which certain methods have been investigated and determine their performance under various scenarios.

METHODS

A systematic search of several electronic databases including MEDLINE and Scopus was carried out from conception to 30th November 2022. Included papers were published in a peer-reviewed journal, readily available in the English language and focused on comparing relevant methods in a superiority randomised controlled trial under a simulation study. Articles were screened using these criteria and a predetermined extraction form used to identify relevant information. A quality assessment appraised the risk of bias in individual studies. Extracted data was synthesised using tables, figures and a narrative summary. Both screening and data extraction were performed by two independent reviewers with disagreements resolved by consensus.

RESULTS

Of 2325 papers identified, 267 full texts were screened and 17 studies finally included. Twelve methods were identified across papers. Instrumental variable methods were commonly considered, but many authors found them to be biased in some settings. Non-compliance was generally assumed to be all-or-nothing and only occurring in the intervention group, although some methods considered it as time-varying. Simulation studies commonly varied the level and type of non-compliance and factors such as effect size and strength of confounding. The quality of papers was generally good, although some lacked detail and justification. Therefore, their conclusions were deemed to be less reliable.

CONCLUSIONS

It is common for papers to consider instrumental variable methods but more studies are needed that consider G-methods and compare a wide range of methods in realistic scenarios. It is difficult to make conclusions about the best method to deal with non-compliance due to a limited body of evidence and the difficulty in combining results from independent simulation studies.

PROSPERO REGISTRATION NUMBER

CRD42022370910.

Toward a rigorous assessment of the statistical performances of methods to estimate the Minimal Important Difference of Patient-Reported Outcomes: a protocol for a large-scale simulation study

Interpreting observed changes over time in Patient-Reported Outcomes (PRO) measures is still considered a challenge. Indeed, concluding an observed change at group level is statistically significant does not necessarily equate this change is meaningful from the perspective of the patient. To help interpret within and/or between group changes in the measure over time, the estimation of the Minimal Important Difference (MID) of the instrument - the smallest value that patients consider as a perceived change - is useful. In the last 30 years, a plethora of methods and estimators have been proposed to derive this MID value using clinical data from sample of patients. MIDs for hundreds of PROs have been estimated, with frequently a substantial variability in the results depending on the method used. Nonetheless, a rigorous assessment of the statistical performances of numerous proposed methods for estimating MIDs by experimental design such as Monte-Carlo study has never been performed. The purpose of this paper is to thoroughly depict a protocol for a large-scale simulation study designed to investigate the statistical performances, especially bias against a true populational value, of the common proposed estimators for MID. This paper depicts how investigated methods and estimators were retained after the conduct of a systematic review, the design of a conceptual model that formally defines what is the true populational MID value and the translation of the conceptual model into a model allowing the simulation of responses of items to a hypothetical PRO at two times of measurement along with the response to a Patient Global Rating of Change at the second time under the constraint of a known true MID value. A statistical analysis plan is depicted in order to conclude if working hypotheses on what could be appropriate MID estimators will be verified. Strengths, assumptions, and limits of the simulation model are exposed. Finally, we show how this protocol could be the basis for fostering future methodological research on the issue of interpreting changes in PRO measures.

Detection of response shift in health-related quality of life studies: a systematic review

Background

Response Shift (RS) refers to the idea that an individual may undergo changes in its health-related quality of life (HRQOL). If internal standard, values, or reconceptualization of HRQOL change over time, then answer to the same items by the same individuals may not be comparable over time. Traditional measures to evaluate RS is prone to bias and strong methodologies to study the existence of this phenomenon is required. The objective is to systematically identify, analyze, and synthesize the existing and recent evidence of statistical methods used for RS detection in HRQOL studies.

Methods

The analysis of selected studies between January 2010 and July 2020 was performed through a systematic review in MEDLINE/PubMed, Scopus, Web of Science, PsycINFO and Google Scholar databases. The search strategy used the terms “Health-Related Quality of Life” and “Response Shift” using the filters “Humans”, “Journal Article”, “English” and “2010/01/01–2020/07/31”. The search was made in August 2020.

Results

After considering the inclusion and exclusion criteria, from the total selected articles (675), 107 (15.9%) of the publications were included in the analysis. From these, 79 (71.0%) detected the existence of RS and 86 (80.4%) only used one detection method. The most used methods were Then Test (n = 41) and Oort’s Structural Equation Models (SEM) (n = 35). Other method used were Multiple Lineal Regression (n = 7), Mixed-Effect Regression (n = 6), Latent Trajectory Analysis (n = 6), Item Response Theory (n = 6), Logistics Regression (n = 5), Regression and Classification Trees (n = 4) and Relative Importance Method (n = 4). Most of these detected recalibration, including Then Test (n = 27), followed by Oort’s SEM that detected the higher combination of RS types: recalibration (n = 24), reprioritization (n = 13) and reconceptualization (n = 7).

Conclusions

There is a continuous interest of studying RS detection. Oort’s SEM becomes the most versatile method in its capability for detecting RS in all different types. Despite results from previous systematic reviews, same methods have been used during the last years. We observed the need to explore other alternative methods allowing same detection capacity with robust and highly precise methodology. The investigation on RS detection and types requires more study, therefore new opportunity grows to continue attending this phenomenon through a multidisciplinary perspective.

Critical examination of current response shift methods and proposal for advancing new methods

Purpose

This work is part of an international, interdisciplinary initiative to synthesize research on response shift in results of patient-reported outcome measures. The objective is to critically examine current response shift methods. We additionally propose advancing new methods that address the limitations of extant methods.

Methods

Based on literature reviews, this critical examination comprises design-based, qualitative, individualized, and preference-based methods, latent variable models, and other statistical methods. We critically appraised their definition, operationalization, the type of response shift they can detect, whether they can adjust for and explain response shift, their assumptions, and alternative explanations. Overall limitations requiring new methods were identified.

Results

We examined 11 methods that aim to operationalize response shift, by assessing change in the meaning of one’s self-evaluation. Six of these methods distinguish between change in observed measurements (observed change) and change in the construct that was intended to be measured (target change). The methods use either (sub)group-based or individual-level analysis, or a combination. All methods have underlying assumptions to be met and alternative explanations for the inferred response shift effects. We highlighted the need to address the interpretation of the results as response shift and proposed advancing new methods handling individual variation in change over time and multiple time points.

Conclusion

No single response shift method is optimal; each method has strengths and limitations. Additionally, extra steps need to be taken to correctly interpret the results. Advancing new methods and conducting computer simulation studies that compare methods are recommended to move response shift research forward.

Supplementary Information

The online version of this article (10.1007/s11136-020-02755-4) contains supplementary material, which is available to authorized users.