The data that runs pharma: trials to real-world evidence

Each phase is a gate. Most compounds fail. Industry estimates commonly suggest only a small fraction of drugs entering Phase 1 ever reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.Voir la définition complète → approval, though the exact rate varies by disease area.

What makes trial data special

Trial data is prospective (planned before collection) and structured. Before a single patient enrolls, the sponsor writes a protocol and a statistical analysis plan that define the primary endpoint (the main outcome, for example "overall survival at 12 months") and exactly how it will be measured and analyzed. You cannot go fishing for a good result after the fact.

This discipline is enforced by regulators. In the US that is the FDA (Food and Drug Administration); in the EU, the EMA (European Medicines Agency). Trials must follow Good Clinical Practice (GCP), an international quality standard. Every data point must be traceable back to a source document, a principle summarized by the acronym ALCOA: data should be Attributable, Legible, Contemporaneous, Original, and Accurate.

Public trial registration adds transparency. Anyone can browse ongoing and completed studies at ClinicalTrials.gov, a free registry that includes protocol details and, increasingly, results.

The limits of the trial

RCTs are powerful precisely because they are artificial. And that artificiality is also their weakness.

Trials enroll carefully selected patients: specific age ranges, few other illnesses, good adherence. Real patients are messier. They are older, take five other medications, skip doses, and have conditions the trial excluded.

Trials are also short and narrow. A two-year trial cannot detect a side effect that appears after five years, or one that occurs in one patient per hundred thousand. And trials rarely answer practical questions like "does this work as well as the three cheaper drugs already on the market?"

This gap is where the second data stream enters.

Real-world dataReal-world dataRWD, données collectées en dehors des essais cliniques contrôlés : dossiers médicaux, claims d'assurance, données de dispositifs connectés, base des Real-World Evidence (RWE). and real-world evidence

Real-world data (RWD) is health data collected during routine care, not in an experiment. The main sources:

• Electronic health records (EHRs): clinical notes, diagnoses, lab results, prescriptions from hospitals and clinics.
• Claims data: billing records from insurers and payers. Excellent for tracking what was prescribed, dispensed, and paid for across large populations.
• Registries: databases tracking all patients with a specific disease or device.
• Pharmacy and increasingly wearable or app data.

Real-world evidence (RWE) is the clinical conclusion you derive from analyzing RWD. The distinction matters: raw claims records are data; a study showing a drug reduces hospitalizations in everyday practice is evidence.

The trade-off

RWD is vast, cheap to access, and reflects actual patients. But it is observational, not randomized. Two groups taking different drugs may differ in a hundred hidden ways (sicker patients get the stronger drug, for example). This is confounding, and it is the central challenge of RWE.

Analysts fight confounding with statistical methods. A common one is propensity score matching: pairing treated and untreated patients who look similar across measured characteristics, to approximate a randomized comparison.

Here is the intuition in a few lines of pseudocode:

The catch: this only balances what you can measure. Unmeasured differences still bias the result. That is why RWE is powerful but rarely accepted as standalone proof of efficacy for a brand-new drug.

How the two streams combine across a drug's life

The clinical and real-world streams are not rivals. They cover different stages.

Before approval: RCT data does the heavy lifting. The FDA or EMA approves based on trial evidence of efficacy and safety. RWD may support the case, for example by defining how a disease progresses naturally (a natural history study) to serve as a comparison for a rare-disease trial where a placebo group would be unethical.

At approval and label expansion: regulators now sometimes accept RWE for specific uses. The FDA has published a framework for evaluating real-world evidence, driven by the 21st Century Cures Act. RWE has been used, for instance, to support new indications for already-approved drugs, where an RCT exists for the original use.

After approval (this is the big one): pharmacovigilance (drug safety monitoring) runs largely on real-world datareal-world dataRWD, données collectées en dehors des essais cliniques contrôlés : dossiers médicaux, claims d'assurance, données de dispositifs connectés, base des Real-World Evidence (RWE).. Phase 3 trials cannot catch rare or long-term harms. Once millions of people take a drug, safety signals emerge in EHRs, claims, and adverse-event reports. Regulators run active surveillance systems (the FDA's Sentinel Initiative queries a large network of real-world datareal-world dataRWD, données collectées en dehors des essais cliniques contrôlés : dossiers médicaux, claims d'assurance, données de dispositifs connectés, base des Real-World Evidence (RWE).) to spot problems fast. A safety signal can trigger a label warning or, rarely, withdrawal.

A concrete illustration

Imagine a new drug for type 2 diabetes.

Phase 3 trial shows it lowers blood sugar (measured by HbA1c, a standard lab marker) better than placebo over one year, in 4,000 relatively healthy enrolled patients. That earns approval.

Post-launch, the drug reaches millions, including elderly patients with kidney disease who were excluded from the trial. Claims and EHR analysis reveals two things the trial never could: a rare but serious side effect appearing in roughly one in fifty thousand users, and evidence that in everyday use, adherence is far lower than in the trial, blunting real benefit.

The trial proved it *can* work under ideal conditions. The real-world evidence shows how it *does* work in practice, and keeps watch on safety. Both are needed. Neither is sufficient alone.

Why the rigor exists

The asymmetry is stark: a drug that helps helps gradually, but a drug that harms can harm catastrophically and irreversibly. History is full of costly lessons where inadequate data let unsafe products through. Regulatory rigor (GCP, ALCOA, prospective protocols, independent review) is the industry's answer to that asymmetry. The paperwork is not bureaucracy for its own sake. It is the mechanism that lets a stranger trust a molecule they will put in their body.

Key Takeaways

• Trial data proves causation; real-world evidence proves relevance. RCTs answer "does it work under controlled conditions?" RWE answers "does it work, and stay safe, in real patients over time?"
• Rigor is structural, not optional. Prospective protocols, defined endpoints, GCP, and ALCOA traceability exist because you cannot un-harm a patient. Data integrity is the drug's license to exist.
• RWD is abundant but confounded. Claims and EHRs reflect real patients cheaply, but lack randomization. Methods like propensity matching help, which is why RWE rarely stands alone as proof of a new drug's efficacy.
• The two streams cover different life stages. Trials dominate before approval; real-world datareal-world dataRWD, données collectées en dehors des essais cliniques contrôlés : dossiers médicaux, claims d'assurance, données de dispositifs connectés, base des Real-World Evidence (RWE). dominates safety monitoring (pharmacovigilance) after launch, and is increasingly used to expand what a drug is approved for.
• Regulators are opening the door to RWE deliberately. Frameworks from the FDA and EMA are defining where real-world evidence is acceptable, but they hold it to explicit standards rather than accepting it wholesale.