SNDS (Systeme National des Donnees de Sante)

SNDS

(Système National des Données de Santé, formerly SNIIRAM) consolidates France's nationwide health-insurance claims — covering essentially the entire resident population (~66 million) — with the PMSI hospital-discharge database and cause-of-death records. Because statutory insurance is universal, every reimbursed physician visit, dispensed prescription (with dispensing dates), imaging study, lab test, and device appears; private-practice and hospital care both capture, making coverage near-total rather than sampled.

Why it matters for RWE

Whole-population capture removes the selection questions inherent in insured-population databases: denominators approximate the true population, enabling incidence rates directly comparable to vital statistics, ultra-rare-event safety surveillance, complete life-course exposure histories, and immigrant/migrant cohort construction impossible elsewhere. France's long history of using these data for public-health regulation (pioneered by ANSM and CépiDc studies) gives an unusually rich methods literature.

Operational characteristics

• Beneficiary files: demographic and registration spans per person; anonymized via permanent pseudonymized identifiers enabling cross-file linkage.
• Care-delivery detail: reimbursed acts coded in CCAM (procedures), prescriptions with dispensing dates and quantity; diagnosis codes on claims are sparse (long-term-disease registrations carry ICD-10), so chronic-condition identification leans heavily on the ALD (affection de longue durée) registry and medication proxies.
• PMSI: rich hospital ICD-10-coded diagnoses/procedures across public/private hospitals — the primary outcome source.
• Access: governed by the Health Data Hub / CNAM processes; project-based approval, remote secure environments.

Common pitfalls

• Sparse outpatient diagnoses: unlike UK GP data, ambulatory diagnoses appear mainly as reimbursement codes and ALD registrations; phenotype algorithms must be medication/act-based more than code-based.
• OTC and non-reimbursed items invisible: low-cost generics fully reimbursed, but some products escape.
• ALD registration lag: registering a long-term illness triggers 100% coverage but administrative timing introduces delay relative to clinical onset.
• Cross-border care missing: treatment outside France does not appear.

Pros, cons, and trade-offs

• vs CPRD: total-population claims completeness vs primary-care clinical depth (BMI, smoking, labs).
• vs US claims: true population denominator vs insured-subset selection; no cost-to-payer granularity comparable to US paid amounts.
• Trade-off: breadth vs depth — SNDS maximizes completeness of utilization; clinical nuance requires PMSI linkage and creative algorithms.

When NOT to use

Questions requiring ambulatory clinical measurements without registry linkage; non-French transportability claims without adjustment; cost-effectiveness needing itemized unit costs beyond tariff schedules.

Decision diagram

flowchart LR
  INS[National insurance funds] --> CL[Reimbursed-care claims]
  H[Hospitals public + private] --> PM[PMSI discharges]
  DC[Cause-of-death registry] --> LK[Pseudonymized linkage] --> SNDS[SNDS]
  ALD[Long-term-disease registrations] --> SNDS
  SNDS --> HDH[Health Data Hub secure access]
  HDH --> S[Whole-population RWE]

Worked example

Scenario

Whole-population incidence of myocarditis after mRNA COVID-19 vaccination by age and sex, using nationwide dispensing and PMSI outcomes.

Dataset

Vaccination-attributable myocarditis incidence estimates.

Steps

Result

Elevated myocarditis SIR concentrated in young males after second dose (SIR ~3.4), consistent with international findings; SCCS confirmed within-person association.

Trade-offs

Runnable example

\
import pandas as pd
import statsmodels.api as sm

def sccs_events(events, vacc_dates, risk_window=(0, 28), control_window=(50, None)):
    # Person-period construction: risk vs control intervals per vaccinated case
    rows = []
    for pid, vd in vacc_dates.items():
        ev = events.get(pid, [])
        for e in ev:
            rows.append({"pid": pid,
                         "risk": int(risk_window[0] <= (e - vd).days <= risk_window[1]),
                         "exposure_period": True})
    df = pd.DataFrame(rows)
    # Conditional Poisson approximation via logit with pid fixed effects
    X = pd.get_dummies(df.risk, drop_first=True)
    res = sm.GLM(df.risk, sm.add_constant(X),
                 family=sm.families.Poisson()).fit()
    return res

\
library(SCCS)

fit_sccs <- function(events_df, vacc_date, risk_start = 0, risk_end = 28) {
  # events_df: one row per event with indiv (person id) and event date
  sca <- sccs_data.frame(indiv = events_df$indiv,
                         event = events_df$event_date,
                         exposure = setNames(vacc_date, events_df$indiv))
  mod <- standardsccs(event ~ risk(risk_start:risk_end), data = sca)
  summary(mod)
}

\
proc genmod data=person_periods;
  class pid risk(ref='0');
  model events = risk / dist=poisson link=log offset=log_days;
  repeated subject=pid / type=ind;
  estimate 'Risk ratio' risk 1 -1 / exp;
run;

Citations

1. [1]Tuppin P, et al. Value of a national administrative database to guide public decisions: From the système national d'information interrégimes de l'Assurance Maladie (SNIIRAM) to the système national des données de santé (SNDS) in France. Revue d'Épidémiologie et de Santé Publique. 2017.
2. [2]Health Data Hub. Le Système National des Données de Santé (SNDS).

1. [3]Strom BL. Data validity issues in using claims data. Pharmacoepidemiology and Drug Safety. 2001.

1. [4]Santé publique France. Vaccine-safety and epidemiologic surveillance publications using SNDS.