Article Text
Statistics from Altmetric.com
We read with great interest the manuscript published by Li and colleagues that was published in the Annals of the Rheumatic Diseases in 2020.1 They evaluated the associations of regular glucosamine use with all-cause and cause-specific mortality in a large prospective cohort. This study provides valuable and interesting results but some methodological concerns should be taken into account.
First, they presented the results in term of two models (model 1 and model 2), but it is not clear how the models were built. The rationale for the confounder selection was not provided. Causal diagrams (directed acyclic graphs) are a new approach used in the epidemiological literature to conceptualise confounding effects and to identify minimal sufficient adjustment sets.2 Lin and colleagues have not explained these steps in their causal study and all baseline variables have been included in their multivariable models.3
Second, the authors constructed propensity scores using all baseline covariates,1 but whether or not confounder variables are distributed equally between the glucosamine users and non-users appears not to have been examined. Standardised mean difference is the most commonly used statistic for examining the balance of confounding variables between groups when propensity scores are applied in a study. In fact, the success of propensity score modelling should be judged using the balance of confounders between the glucosamine user and non-users. In addition, propensity scores could be included into the model as a covariate to adjust for baseline differences. However, the assumption regarding the functional relationship between the propensity scores and outcomes (linearity, proportional hazards, etc) needs to be assessed to avoid any biases estimates.4
Third, they indicated that the proportional hazard assumption was evaluated in their study, but the results of the statistical test and hazard curves were not reported.
Fourth, while the baseline characteristics of glucosamine users and non-users are presented in their Table 1, p values were not reported so it is unclear whether the differences between groups are statistically significant.
Finally, the reasons for loss to follow-up are not reported and it is not clear how this important issue was handled in their longitudinal study. Both differential and non-differential reasons for loss to follow-up need to be considered, and differential reasons can lead to selection bias.5
Ethics statements
Patient consent for publication
Footnotes
Contributors SS and MAM wrote the initial manuscript and finalised it.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; internally peer reviewed.