HSR IRCCS Statistical Analyses

This repository presents a small portfolio of clinical statistical analyses developed for research work connected with IRCCS Ospedale San Raffaele (HSR), Milan. HSR is part of Gruppo San Donato.

The notebooks focus on oral surgery, implantology, peri-implant outcomes, and medically complex patient cohorts. They are written as reproducible analysis notebooks, with emphasis on transparent data validation, appropriate statistical testing, model interpretation, and clear reporting of limitations.

This is a portfolio/research-code repository, not an official institutional publication. Any clinical interpretation should be read in the context of the corresponding study protocol, manuscript, and available dataset.

Notebooks

Notebook	Topic	Statistical focus
mouthwash_mixed_effects.ipynb	Comparison of Euchlorine and Chlorhexidine after oral surgery, using Plaque Index and Bleeding on Probing across T0, T1, and T2.	Linear mixed-effects models with patient-level random intercepts, time-by-treatment interaction tests, adjusted sensitivity models, and FDR-controlled subgroup analyses.
ligosan_wilcoxon_mixed.ipynb	Six-month LIGOSAN periodontal outcome analysis, including PPD, BoP, PI, and suppuration in test and control groups.	Baseline comparability checks, Shapiro-Wilk diagnostics, paired Wilcoxon tests, adjusted OLS models, mixed-effects repeated-measures models, and McNemar tests for paired binary outcomes.
hiv_implant_km_cox.ipynb	Implant survival analysis in an HIV-related clinical cohort, with survival, hazard, and marginal bone loss summaries.	Kaplan-Meier survival estimation, log-rank testing, Cox proportional hazards regression, Schoenfeld-residual checks, and bootstrap sensitivity analysis for five-year survival differences.
cvd_implant_km_mbl.ipynb	Seven-year full-arch implant rehabilitation follow-up in patients with cardiovascular disease versus controls.	Reconstructed Kaplan-Meier survival analysis, log-rank testing, marginal bone loss trajectory visualization, published-result reconciliation, exploratory Cox-model interpretation, and power sensitivity analysis.
graft_prepost_robust_ols.ipynb	Palatal versus tuber graft analysis, covering baseline balance, post-surgical clinical outcomes, satisfaction, and marginal bone loss.	Welch and Fisher baseline tests, Mann-Whitney sensitivity checks, McNemar and Wilcoxon paired pre/post tests, FDR adjustment, and robust HC3 OLS models for change scores.

Repository Scope

The analyses show how clinical research questions were translated into auditable statistical workflows:

• data loading and schema validation before inference;
• descriptive summaries for cohort composition and baseline balance;
• explicit choice of parametric, non-parametric, paired, and survival methods;
• effect-size reporting where useful;
• multiplicity adjustment for exploratory endpoint sets;
• sensitivity analyses for small samples, sparse events, or model assumptions;
• plain-language interpretation of the statistical outputs.

Data

The notebooks expect local datasets under data/. Clinical source data are intentionally excluded from version control through .gitignore, because the repository is intended for displaying analysis structure and statistical methodology rather than distributing patient-level data.

Some notebooks also generate figures in images/ when executed. These outputs can be regenerated from the notebooks when the required local data are available.

Typical Environment

The analyses use Python notebooks with the scientific Python stack:

• pandas, numpy
• scipy
• statsmodels
• matplotlib, seaborn
• jupyter

Install the dependencies in a local environment, place the required private datasets in data/, then run the notebooks from the repository root so their relative paths resolve correctly.