We thank the authors for their interest and will clarify the issues raised in the letter [1]. The authors write that “the criteria employed for variable selection remain undisclosed”; however, we depicted the reasoning behind our choice of confounders in the directed acyclic graph (DAG) which accompanied the study (Figure S1). This DAG clarifies which confounders were adjusted for in accordance with the STROBE guidelines. The paths in the DAG show the flow of causality and bias and were used to guide the subsequent statistical analysis [2, 3].
We used the classical definition of a confounder: a variable associated with the exposure, associated with the outcome, and not on the causal pathway (not an intermediate). Marital status, infant sex and household income can influence pregnancy outcomes, but these are not (based on the literature) associated with the risk of developing maternal myasthenia gravis (MG). In the DAG, we showed that marital status (cohabitation) influences pregnancy outcomes but not maternal MG. Infant sex does not influence maternal MG. Arguably, the chronic illness MG could affect household income, which could affect pregnancy outcomes. Still, household income would be an intermediate on the causal pathway and not a true confounder. One should not adjust for intermediates in the logistic regression model because the “intermediate path” is open and is part of the total effect of the exposure on the outcome.
The authors suggest using various statistical methods to tell us which risk factors to include as confounders. In fact, the suggested methods are known to introduce bias [4, 5]. One must distinguish between etiological research and prediction research. In etiological research, data‐driven analyses do not work to select confounders; we need causal information from outside the data based on prior biological knowledge. The aim of this study is causal inference (etiological research) and not prediction. Our research question was to investigate if there is a causal association between maternal MG and adverse pregnancy outcomes. We used the DAG to display our prior biological knowledge. As confounders are based on biological knowledge, they are interesting to debate.
Regarding the suggestion of using a pseudo R2, we do not see the relevance. Model fit of the statistical model is crucial in prediction research but not important in etiological research. For etiological research, we include variables if they are considered confounders by the DAG. Pseudo R2 and other goodness of fit measures are mostly useful in prediction modeling [6].
Finally, we agree that maternal use of MG medications may influence neonatal outcomes; however, this was not our research question. We did not find any higher rates of congenital malformations in the infants of mothers with MG. We only had data on MG medications taken during pregnancy and not before conception. Mycophenolate mofetil was not used during pregnancy by any of the women with MG. We agree that medication is a topic of great interest to patients and physicians and our next paper will look in greater detail at maternal use of MG medications during pregnancy.
AUTHOR CONTRIBUTIONS
Anna Rostedt Punga: Conceptualization; methodology; formal analysis; supervision; writing – review and editing. Laura O'Connor: Methodology; investigation; formal analysis; writing – original draft. Anna‐Karin Wikström: Conceptualization; methodology; formal analysis; supervision; writing – review and editing.
CONFLICT OF INTEREST STATEMENT
The authors report no financial or other conflicts of interest.
ACKNOWLEDGEMENTS
The authors thank statistician Katja Gabrysch at the Uppsala Clinical Research Center for assistance and discussion regarding the statistical analysis.
EJoN‐23‐1134: Pregnancy outcomes for women with myasthenia gravis and their newborns: a nationwide register‐based cohort study
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.
REFERENCES
- 1. O'Connor L, Malmeström C, Da Silva Rodrigues R, Brauner S, Wikström AK, Punga AR. Pregnancy outcomes for women with myasthenia gravis and their newborns: A nationwide register‐based cohort study. Eur J Neurol. 2023;31(1):e16100. doi: 10.1111/ene.16100. Epub ahead of print. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Textor J, Hardt J, Knüppel S. DAGitty: a graphical tool for analyzing causal diagrams. Epidemiology. 2011;22(5):745. [DOI] [PubMed] [Google Scholar]
- 3. Shrier I, Platt RW. Reducing bias through directed acyclic graphs. BMC Med Res Methodol. 2008;8:70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Heinze G, Dunkler D. Five myths about variable selection. Transpl Int. 2017;30(1):6‐10. [DOI] [PubMed] [Google Scholar]
- 5. Greenland S, Pearce N. Statistical foundations for model‐based adjustments. Annu Rev Public Health. 2015;36:89‐108. [DOI] [PubMed] [Google Scholar]
- 6. Petrie A. Medical Statistics at a Glance Fourth Edition. Wiley Blackwell; 2020:82. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.