R statistical software

Baseline characteristics are presented as numbers and proportions. Comparisons between categorical variables were tested using chi-square tests. This retrospective cohort is divided into four groups according to ART and GDM: non-ART mothers without GDM, ART mothers without GDM, non-ART mothers with GDM, and ART mothers with GDM. Logistic regression models were used to estimate odds ratios (ORs) and 95% confidence intervals (95% CIs) for PTB, after adjusting for potential confounders, including maternal age, race/ethnicity, education, marital status, parity, pre-pregnancy body mass index, hypertension before pregnancy, previous history of PTB, smoking before pregnancy, smoking during pregnancy, initiation of prenatal care, gestational hypertension or preeclampsia & eclampsia and infant sex. We further conducted secondary analyses stratified by maternal age (< 25, 25-34, ≥ 35 years old). P for interaction was calculated on the basis of multivariable logistic regression models with multiplicative interaction term (age/pre-pregnancy body mass index *exposure groups). All analyses were performed with R statistical software (version 3.6.4) and GraphPad Prism 8 (GraphPad Software, San Diego, CA). A two-sided P < 0.05 was considered statistically significant.

Prism software, version 7 (GraphPad Software, Inc., San Diego, CA, USA) and R statistical software [70 ] were used for statistical analysis. Non-parametric tests were used—the Mann–Whitney U test for the two-group comparisons, the Kruskal–Wallis test for multiple-group comparisons, and the Spearmen rank test for correlations between sCD14 and CD4 counts or CRP values. The WHO, CDC, and other major TB stakeholders recently recommended a sensitivity of at least 90% with a specificity of 70–80% for a non-sputum-based diagnostic POC TB [14 (link)]. We therefore chose a cutoff for a positive sCD14 value based on ROC curve analysis, with the objective of obtaining over 90% (ideally 95%) sensitivity with at least 80% specificity using values from the smear-negative culture-confirmed South African TB patients and South African controls.

Statistical analyses were performed using R Statistical Software and Graphpad Prism 8.0. Protein concentrations were displayed as median with interquartile range (IQR). Differences in clinical characteristics were calculated with the Mann-Whitney U test for continuous and Fisher Exact tests for categorical data. Because repeated measures ANOVA cannot handle missing values, and we had different numbers of samples in each cohort, we analysed differences in lectin protein levels from all available samples by fitting a mixed model in Graphpad Prism 8.0. This mixed model uses a compound symmetry covariance matrix and is fitted using Restricted Maximum Likelihood (REML). We adjusted the data for non-sphericity with the Geisser-Greenhouse correction. Differences between first sample lectin pathway concentrations were calculated with Kruskall-Wallis tests, follow-up comparison of the mean rank of every column, and adjustment of P values for multiple comparisons. We calculated correlations by applying Pearson’s tests to log-transformed data that did not include repeat measures, and linear mixed models with a repeated measures correlation technique (rmcorr) to data from COVID-19 cohorts (23 (link)). We adjusted p-values for multiple comparisons using the method of Benjamini and Hochberg with a false discovery rate (Q) of 5% (24 (link)).

Data are shown as mean ± SD. An unpaired Student’s t test with Welch’s correction or one-way ANOVA with Tukey’s multiple comparison test was used to analyze variance among experimental groups. P < 0.05 was considered significant. Data in the figures are annotated as follows: *P < 0.05; **P < 0.01; and ***P < 0.001. Prism software (Graph Pad) or R Statistical Software was used to generate data graphs and perform statistical analyses.

A two-sided P value less than 0.05 was considered statistically significant. The Akaike information criterion (AIC) value was calculated to assess the risk of overfitting (18 ). The deep learning model was developed using Python (version 3.6.7). The CPH regression model and the C statistical were determined by survival, survminer, and rms packages with R statistical software (Version 4.2.0), and the survival curves were plotted using GraphPad Prism 7 (GraphPad Software).