Additional file 1 of Effect of discontinuing antipsychotic medications on the risk of hospitalization in long-term care: a machine learning-based analysis
Bibliographic record
Abstract
Additional file 1: Supplementary methods: additional details on data preprocessing, workflow of causal machine learning (ML) model training and evaluation, assumptions of causal modeling, confounder selection, individual treatment effect models, model evaluation, correlation matrix, performance of ML models for predicting factual outcomes, treatment effect distributions, heterogeneity, model interpretation, and sensitivity analyses. Fig. S1 Treatment definition. Two study groups were defined, treatment (antipsychotic discontinuing group) and control (antipsychotic chronic users) groups from the data of the long-term care residents who had at least four RAI assessments. Resident was classified in the treatment group if antipsychotics were prescribed at the baseline period (assessments 1 and 2) but no at the follow-up period (assessments 3 and 4). Resident was classified in the control group if antipsychotics were prescribed at the baseline and follow-up periods (assessments 1–4). The outcome of hospitalization was measured within 360 days of the first follow-up RAI assessment (assessment 3). The input variables in the models were collected from baseline (assessment 2) before treatment. For residents with more than one group of four RAI assessments, the assessment group was randomly selected from the valid groups. Fig. S2 Distributions of the estimated treatment effect values. Distributions of the estimated individual treatment effect (ITE) values of (a) X-learner, (b) double robust (DR) learner, (c) double machine learning (DML), and (d) causal forest. Fig. S3 Surrogate model of causal forest model. Surrogate decision tree derived from the causal forest model, illustrating key variables contributing to treatment effect heterogeneity. Fig. S4 Confounder balance of antipsychotic discontinuing and chronic user groups. Confounder balance between residents in the two study groups before and after inverse propensity weighting assessed with the absolute standardized mean difference (SMD). Confounders with absolute SMD below 0.1 are considered well balanced. Table S1 Variables that were processed from the source data. List of variables that were processed from the data source and univariate performance for predicting outcome (hospitalization) and treatment (antipsychotic discontinuation). Univariate performance was calculated by tenfold cross-validation of training data with logistic regression algorithm. Parameter “coef” indicates the direction of the effect. Table S2 Confounders. List of confounders that were selected by a group of three study researchers, experts from the field of clinical geriatrics and pharmacology. Table S3 Correlation values of the estimated treatment effects. Correlation values between the individual treatment effect estimates of double machine learning (DML), double robust (DR) learner, X-learner, and causal forest of test data. Table S4 Sensitivity analysis 1. Average treatment effect (ATE), area under uplift curve (AUUC), and c-for-benefit values for double machine learning (DML), double robust (DR) learner, X-learner, and causal forest models. Data randomly split for training and testing sets. Table S5 Sensitivity analysis 2. Average treatment effect (ATE) and 95% confidence intervals (CI) for double machine learning (DML), double robust (DR) learner, X-learner, and causal forest of test data when actual treatment variable or outcome was replaced with a random variable or a random confounder was added in the model.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.071 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.002 | 0.004 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.802 | 0.079 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".