From the datapoint
to the bedside.
I’m Paulina Del Mundo: physician, MPH from Hopkins (epidemiology and biostatistics), clinical data scientist. Every clinical recommendation rests on a chain that runs from a raw measurement up through an estimate, a trial, and a decision rule. This is my working reference for tracing that chain, from the bedside guideline down to the datapoint it rests on, and back up: study design, causal inference, sensitivity analysis, and the methods writing that holds them together.
MD MPH Hopkins 24.6M+ encounters analyzed
The pathway Explore the pathway →
Every recommendation a clinician acts on sits at the end of a pathway. Build it, framing a question and then constructing each rung up to a sound recommendation, or trace it, taking a recommendation that exists and walking back down to the measurement it rests on. Same rungs either way, with a cross-cutting set of moves for defending the result.
New here? The pathway’s Framing rung opens by sorting your question to the kind of evidence it needs, before any method.
The question and the claim the study is built to answer
01 — MeasurementHow the raw datapoint was defined and collected
02 — ModelHow the estimate was produced, and what it assumes
03 — EstimateThe effect size and the uncertainty around it
04 — SynthesisThe trial or meta-analysis it is pinned to
05 — Decision ruleThe threshold or calculator that makes it actionable
06 — RecommendationThe guideline sentence a clinician acts on
∗ — Defend itThe cross-cutting moves that stress-test the whole chain
§ — Conduct itThe ethics, governance, and documents that let a study legitimately run
Traces All traces →
Measurement rung
The 120 mmHg systolic target
The 2017 ACC/AHA intensive target leans on a trial that measured blood pressure differently from most clinics. The trace walks the recommendation down to that measurement choice.
Survey-weighted analysis
NHANES cardiometabolic risk
Survey-weighted analysis, MI, calibration vs AUC, structural-overlap diagnostics in a head-to-head with the Pooled Cohort Equations.
Difference-in-differences
Part D insulin DiD
A clean Inflation Reduction Act natural experiment: TWFE DiD, event-study, parallel-trends defense, placebo and leave-one-out robustness.
Real-world data
Medicaid outlier detection
Robust statistics on heavy-tailed data, BH-FDR multiplicity, isolation-forest second opinion, ontology-aware claims-data analysis.
Model & estimate rungs
Tenecteplase vs alteplase
A non-inferiority appraisal of the 2026 AHA/ASA co-equal thrombolytic recommendation: what a “not worse by more than a margin” result, plus a logistical edge, can and cannot support.
Trial data standards · SAS & R
CDISC clinical-trial programming
Double-programming an FDA-grade analysis package in SAS and R on the CDISC pilot data: SDTM to ADaM, dictionary coding, tables/figures/listings, byte-level reconciliation.
Follow the work
New traces and methods go up regularly. Create a free account → to follow along; everything on the site stays free to read.