Gender

Basic descriptive statistics (mean, median, standard deviation, interquartile range, skewness, and kurtosis) were computed for all variables, which were subsequently tested for normality using the Kolmogorov–Smirnov and Shapiro–Wilk tests. Differences in interval variables (e.g. Hsp90, age, etc.) were tested using the Mann–Whitney U test, while the chi-square test was used to compare frequency counts of categorical variables (e.g. gender). The bivariate relationships between variables under study were assessed using the Spearman correlation coefficient. Linear regression analysis was used to predict patients' Hsp90 levels by a set of predictors (FEV1, FVC, DLCO, SpO₂) while adjusting for a confounder (CRP). Friedman's test was used to analyze repeated longitudinal measurements taken: (a) at baseline, (b) after 1 month, (c) after 6 months, and (d) after 12 months of therapy with cyclophosphamide. Data are presented as median (IQR) unless stated otherwise. Statistical significance was set at p < 0.05. All analyses were conducted using SPSS version 25 (SPSS, Inc., Chicago, IL, USA). Graphs were created using GraphPad Prism 5 (version 5.02; GraphPad Software, La Jolla, CA, USA).

The entire analysis for both microarray and RNA-seq datasets was performed in the R 2.15.0 statistical environment. Samples with missing survival data were excluded from the analysis. Hazard ratio (HR), 95% confidence intervals (CI) and log-rank p values were calculated. We applied the “survival” R package v2.38 (http://CRAN.R-project.org/package=survival/) for Cox regression analysis and the “survplot” R package v0.0.7 (http://www.cbs.dtu.dk/~eklund/survplot/) for generating Kaplan-Meier plots. We determined each percentile of miRNA expression between the lower and upper quartiles of expression as a cutoff point to divide patients into high and low expression groups as described previously^{45 (link)}. Because of the low sample number, a cutoff outside the lower or upper quartile of expression could result in unreliable results. After this, the Cox regression analysis was performed separately for each cutoff. We used the cutoff with the lowest p value as the final cutoff in the Kaplan-Meier analysis. The cutoff values vs. p-values plot was generated to display the overall performance of the selected miRNAs.
In addition, a multivariate survival analysis was performed for the TCGA dataset including – in addition to the miRNAs – clinical data of stage and gender. To plot the distribution of miRNA expression in the high and low expression groups, we applied a one-dimensional scatter plot using the “beeswarm” R package (http://www.cbs.dtu.dk/~eklund/beeswarm/). Correlation between different miRNAs was assessed by computing Spearman rank correlations.

The unprocessed.CEL files were MAS5 normalized in the R environment (http://www.r-project.org) using the simpleaffy library (http://bioinformatics.picr.man.ac.uk/simpleaffy/). We have selected MAS5 for normalization as it ranked among the best normalization methods when contrasted to the results of RT-PCR measurements in our previous study [26] (link). Moreover, MAS5 can be applied to single arrays, enabling seamless future extensions of the database. For the complete database, only the common probes measured in all three array platforms were retained (n = 22,277). Then, a second scaling normalization was performed to center the mean expression for each array to 1000 - this technique can significantly reduce batch effects. Gene expression and clinical data were integrated using PostgreSQL, an open source object-relational database system (http://www.postgresql.org/).
To assess the prognostic value of a gene, each percentile (of expression) between the lower and upper quartiles were computed and the best performing threshold was used as the final cutoff in a univariate Cox regression analysis. Histology, grade, stage, gender and smoking history can be used in the multivariate analysis. However, the multivariate analysis uses less patients as the univariate analysis because not each patients has all clinical information. Kaplan-Meier survival plot and the hazard ratio with 95% confidence intervals and logrank P value were calculated and plotted in R using the “survplot” function of the “survival” Bioconductor package. The R script used by the software to perform the Kaplan-Meier analysis and to identify the best cutoff is available as R script S1.
The entire computational pathway is made accessible for re-analysis in a platform independent online available software running on a Debian Linux (http://www.debian.org) server powered by Apache (http://www.apache.org). The scripts on the server-side were developed in PHP, these control the user interface, the requests and the delivery of the results. The RODBC package provides a middleware layer between R and the PostgreSQL database. This platform can be reached over the internet via http://www.kmplot.com/lung.

Statistical analysis included descriptive statistics for the prevalence of co-occurring psychiatric diagnoses in the total sample, and according to types of SUD diagnoses. We compared the characteristics of patients with - and without COD using proportion tests and independent samples t-tests. The prevalence of types of CODs was examined for the following psychiatric disorders: anxiety (F40-F49); mood (F30-39), ADHD (F90-90.9); personality disorder (F60-69); multiple CODs. Gender differences in the prevalence of each types of CODs were examined using bivariate logistic regression analysis. Bivariate logistic regression analyses were also undertaken to investigate factors associated with relapse. Repeated-measures generalized logistic mixed modeling (GLMM) with a diagonal covariance matrix was used to assess the multivariate association of demographic (age, gender, education), psychological (motivation, mental distress) and types of SUD diagnoses with relapse at 3 month follow-up. The analysis accounted for the prospective nested nature of the data structure (i.e., the same patients nested over time). Since mental distress was measured at two time points (baseline and follow-up), this variable was entered as a time-varying covariate accounting for variation in mental distress across the study period. Variables indicating the center where the patients were treated (unit 1–5) and the length of stay (number of days) were included in the multivariate models to control for any treatment- related differences in relapse rates. We did not incorporate the treatment center variable as a random effect in the analysis due to the small number of patients at each treatment center, which made it complicated to account for the variance of treatment center as a random effect due to the substantial risk of Type II error. The variance inflation factors were < 2 for all independent variables, indicating that multicollinearity was not a concern [35 ]. We ran the GLMM analyses separately for patients with and without COD. SPSS 28 was used for statistical analyses.

Demographic characteristics included age at treatment entry, gender, education level (low: 10 years primary and secondary education or less, or medium/high: high school/vocational school or more) and housing situation (living in owned home/rented housing, or in an unstable living arrangement, including living with family or friends).