4-chlorophenyl methyl sulfide

Figure 1 visualises the simulation design. The simulation procedure generated both Normally distributed continuous predictors and Bernoulli distributed binary predictors, each within clusters of serially correlated variables to represent multiple risk factors that measure similar patient characteristics. Such data were row partitioned into M = 5 distinct subsets of size n_exist = 5000 representing five “existing populations”, and one subset of size n_local representing the “local population”. The M = 5 existing populations were each used to fit an existing logistic regression CPMs representing those available from the literature, with each CPM including a potentially overlapping subset of risk predictors (see Additional file 1: Table S1 for details of predictor selection for the existing CPMs). The single local population was randomly split into a training and validation set, of sizes n_train and n_validate, respectively (i.e. n_local = n_train + n_validate). The training set was used for model aggregation using SR, PCA and PLS in addition to redevelopment using AIC and ridge regression. Datasets frequently only collect a subset of the potential risk factors; to recognise this, exactly those predictors that were included in any of the five existing CPMs were considered candidates during redevelopment. Between simulations n_train was varied through (150, 250, 500, 1000, 5000, 10000); the validation set was reserved only to validate the models with n_validate fixed at 5000 observations. Whilst it is unlikely that local populations would have access to such a large validation set, this was selected here to give sufficient event numbers for an accurate assessment of model performance [21 (link)–23 (link)]. Additionally, although bootstrapping methods are preferable to assess model performance in real-world datasets, the split-sample method was employed here for simplicity and clear illustration of the methods [24 ].Fig. 1

Simulation Procedure: A pictorial representation of the simulation procedure for a given value of population heterogeneity, σ, and a given development sample size, n_train. This process was then repeated across all combinations of σ and n_train

Binary responses were simulated in all populations with probability calculated from a population-specific generating logistic regression model, which included a subset of the simulated risk predictors. The coefficients of each population-specific generating model were sampled from a normal distribution, with a common mean across populations and variance σ. Here, higher values of σ induced greater differences in predictor effects across populations and thus represented increasing between-population-heterogeneity. For each of the aforementioned values of n_train, simulations were run with σ values of (0, 0.125, 0.25, 0.375, 0.5, 0.75, 1).
Across every combination of σ and n_train, the simulation was repeated over 1000 iterations as a compromise between estimator accuracy and computational time. The simulations were implemented using R version 3.2.5 [25 ]. The following packages were used in the simulation: “pROC” [26 (link)] to calculate the AUC of each model, “plsRglm” [27 ] to fit the PLS models and the “cv.glmnet” function within the “glmnet” package for deriving a new model by cross-validated ridge regression [28 (link)]. The authors wrote all other code, which is available in Additional file 1.

Throughout this study, all CPMs will be assumed to be logistic regression models, although the techniques apply to other types of prediction model, such as those for time-to-event outcomes. Stacked regression (SR) [9 (link), 17 ], principal component analysis (PCA) [19 (link), 20 (link)] and partial least squares (PLS) are three possible methods that simultaneously aggregate and calibrate existing models to a new population. We describe SR and PCA here, with PLS described in Additional file 1. This study compares the three aforementioned aggregate approaches with deriving a new model; possible techniques of redevelopment are also outlined in this section.

We used limma-trend version 3.34.8 (Law et al., 2014 (link)) based on the application of this method to scRNA data from Soneson and Robinson (2018) (link). For this analysis, we considered a set of 6,337 genes that had at least 3 unique molecular identifiers (UMIs) in at least 100 cells across all samples. UMI count data are converted to log-scaled counts per million (log-CPM) with an offset of 1. Linear models are fit to the log-CPM profiles of each transcript with membership of each cluster as a binary covariate. To evaluate differential gene expression between clusters, the mean log-CPM of each gene in the cells in each cluster is compared with the mean log-CPM of the gene in all other cells. To evaluate the differential expression of transcripts in each tissue on a percluster basis, we fit independent linear models for the cells in each cluster with tissue membership as a binary covariate. For each transcript, the percluster mean log-CPM for each healthy tissue was compared with the mean across the other two healthy tissues. The mean log-CPMs in psoriatic cells were compared with the means in truncal cells. Moderated t statistics for the differences in means are calculated using an empirical Bayes approach. False discovery rate is calculated from p values associated with the t statistics to evaluate statistical significance.

The textual contents of the two existing PDAs (Ankolekar et al., 2019 (link); PATIENT+, 2020 ; van Tol-Geerdink et al., 2013 (link)) were merged. Alongside this process, survivors (group 1a) were consulted with specific questions about preferences regarding the time frame of side effects (i.e. one or two year after diagnosis) and proxies of side effects (i.e. reliability or stiffness of an erection). Responses were collected via email, stored in a log file and summarized. Moreover, CPMs were created based on static national data from ProZIB (Vernooij et al., 2020 (link)) in order to predict personalized risks on side effects. Key information on these CPMs is provided in Box 1, though more details are published elsewhere (Hasannejadasl et al., 2022a , Hasannejadasl et al., 2022b ).Box 1

Unlabelled Table

CPMs for treatment side effects were based on data from ProZIB (2014–2019), an initiative from the Netherlands Comprehensive Cancer Organization, the professional group for urologists and the prostate cancer patient organization. The database contains demographic information, clinical data and PROMS of patients diagnosed with prostate cancer on three different points in time (upon diagnosis, after one year, after two years). For creating CPMs, the answers on the Expanded Prostate Cancer Index Composite (EPIC26) of 964 patients from 69 hospitals were used. Logistic regression with Recursive Feature Elimination (RFE) was applied and sensitivity, specificity and overall accuracy were calculated.For erection problems, the one-year model required 10 variables (sensitivity 87.1 %; specificity 66.2 %; overall accuracy 75.3 %; AUC 0.84) and the two-year model 9 variables (sensitivity 87.4 %; specificity 62.1 %; overall accuracy 73.7 %; AUC 0.84). Regarding urinary problems, the one-year model was based on 9 variables (sensitivity 82.0 %; specificity 65.1 %; overall accuracy 76.0 %; AUC 0.80) and the two-year model on 9 variables (sensitivity 72.0 %; specificity 40.0 %; overall accuracy 76.0 %; AUC 0.62). Some variables were overlapping; others were different in the one-year and two-year models.

Alt-text: Box 1

A university of applied sciences, an institute for radiation therapy, a provider of PDAs, and two hospitals collaborated in the PROSPECT (prostate cancer decision aid for side effects) project. Deciding upon the most appropriate treatment (RP, EBRT, BT or AS) for an individual patient after being diagnosed with localized prostate cancer was the focus of this PDA. Other PDA characteristics that were formulated in advance:•

an online tool based on the textual content of two existing PDAs;

•

that included personalized risks on side effects based on CPMs;

•

that was introduced to the patient by the urologist or nurse after diagnosis;

•

that could be used by the patient at home to reconsider options;

•

of which the results were then discussed in consultation with the urologist.

Patients with multiple sclerosis (PwMS) and HC underwent a neurological and neuropsychological assessment including the oral version of the Symbol Digit Modalities Test (SDMT) (Smith, 1968 ) and MUSIC test (Kalbe et al., 2013 (link)). Within the MUSIC test three questions on fatigue; subdivided into cognitive and physical fatigue and fatigue impacting work and social life were asked. We examined the HADS (Hospital Anxiety and Depression Scale) questionnaire that were collected at the time of the cognitive examination in our cohort of pwMS (Zigmond and Snaith, 1983 (link)). We then calculated whether symptoms of depression, anxiety and fatigue differed between cognitively preserved and cognitively impaired MS patients using the Mann-Whitney U-Test. SDMT raw scores (Kurtzke, 1983 (link)) were used for the statistical analysis. Clinical disability in pwMS was assessed with the Expanded Disability Status Scale (EDSS) (Kurtzke, 1983 (link)). PwMS were divided into cognitively impaired (CIMS) and cognitively preserved patients (CPMS) using the SDMT z-score at −1.5 standard deviations (Lorefice et al., 2021 (link)). PwMS and HCs were asked the number of years of educations in the neuropsychological assessment.