Background: Statistical model building requires selection of variables for a model depending on the model’s aim. In descriptive and explanatory models, a common recommendation often met in the literature is to include all variables in the model which are assumed or known to be associated with the outcome independent of their identification with data driven selection procedures. An open question is, how reliable this assumed “background knowledge” truly is. In fact, “known” predictors might be findings from preceding studies which may also have employed inappropriate model building strategies. Methods: We conducted a simulation study assessing the influence of treating variables as “known predictors” in model building when in fact this knowledge resulting from preceding studies might be insufficient. Within randomly generated preceding study data sets, model building with variable selection was conducted. A variable was subsequently considered as a “known” predictor if a predefined number of preceding studies identified it as relevant. Results: Even if several preceding studies identified a variable as a “true” predictor, this classification is often false positive. Moreover, variables not identified might still be truly predictive. This especially holds true if the preceding studies employed inappropriate selection methods such as univariable selection. Conclusions: The source of “background knowledge” should be evaluated with care. Knowledge generated on preceding studies can cause misspecification.
There is an increasing interest in machine learning (ML) algorithms for predicting patient outcomes, as these methods are designed to automatically discover complex data patterns. For example, the random forest (RF) algorithm is designed to identify relevant predictor variables out of a large set of candidates. In addition, researchers may also use external information for variable selection to improve model interpretability and variable selection accuracy, thereby prediction quality. However, it is unclear to which extent, if at all, RF and ML methods may benefit from external information. In this paper, we examine the usefulness of external information from prior variable selection studies that used traditional statistical modeling approaches such as the Lasso, or suboptimal methods such as univariate selection. We conducted a plasmode simulation study based on subsampling a data set from a pharmacoepidemiologic study with nearly 200,000 individuals, two binary outcomes and 1152 candidate predictor (mainly sparse binary) variables. When the scope of candidate predictors was reduced based on external knowledge RF models achieved better calibration, that is, better agreement of predictions and observed outcome rates. However, prediction quality measured by cross-entropy, AUROC or the Brier score did not improve. We recommend appraising the methodological quality of studies that serve as an external information source for future prediction model development.
Limited research exists on pregnant women’s knowledge, attitudes, and behavior concerning COVID-19 in sub-Saharan Africa. We performed a cross-sectional study among 648 pregnant women in Fort Portal, Uganda, after the first lockdown starting in June 2020. Structured interviews were conducted at three different facilities during routine antenatal care, assessing sociodemographic background, knowledge of COVID-19, prevention behavior adherence, and psycho-emotional stress levels. We performed descriptive analyses and examined associated factors using multivariable logistic regression. In Fort Portal Region, 32.8% of pregnant women had a higher knowledge regarding the COVID-19 pandemic, while all women at least heard of COVID-19. 88.6% of the women showed low self-reported prevention behavior adherence. More than one third of the pregnant women experienced high psycho-emotional stress related to the pandemic (39.8%). The odds for psycho-emotional stress were increased among the age group 21–30 years (AOR 1.97; 95% CI 1.18–3.35) compared to women under the age of 21, and decreased in single or divorced women compared to women in partnerships (AOR 0.42; 0.22–0.77) and in women having less COVID-19-related knowledge (AOR 0.40; 0.27–0.58). In conclusion, prevention behavior adherence seemed challenging, and psycho-emotional stress was ubiquitous among our cohort. To avoid adverse consequences in maternal and neonatal health, campaigns for hygiene but also women’s emotional state should be a major focus of community healthcare in exceptional times such as the SARS-CoV-2 pandemic.
17 January 2026 in BMC Medical Research Methodology
Background: Missing values occur in almost all real-world medical data. Sometimes, more information is available for the missing values due to technical measurement limits. This was also the case for some sports medical data set where several laboratory measurements below a lower limit of quantification (LLOQ) were faced and supposed to be used in a multiple linear regression model. When studying the literature, the problem arises in several disciplines (environmental epidemiology, pharmacokinetic studies etc.) and different statistical methods are suggested. However, only very limited work on a method comparison is available, especially in the multivariable linear regression settting. Methods: Therefore, we compare statistical methods for addressing values below a LLOQ in multiple linear regression modeling by a simulation study. We consider both the case that the variable below the LLOQ is among one of the independent variables and that it is the dependent variable in the regression model. We also vary different underlying assumptions, such as distributions, sample sizes, proportions of missing values, correlations, or linearity assumptions. Results: Overall, the two compartment model showed the best performance in terms of bias and coverage when the LLOQ occurred in the independent variable and no big collinearity issue was present. When the variable subject to the LLOQ is the dependent variable, tobit showed the lowest bias and highest coverage for censoring proportions up to 0.8. Conclusion: When facing a data set with values below a lower limit of quantification and a multiple linear regression model is chosen as analysis model, a conscious choice for dealing with those left-censored data should be made. In this article, we provide guidance on the performance of different established methods.