Optum Clinformatics Data Mart

Optum's Clinformatics Data Mart

is a de-identified longitudinal US claims database covering commercial, Medicare Advantage, and Medicaid lines - roughly 60-70 million unique members historically with 15-20 million in any year. Medical claims (ICD-10-CM/CPT/HCPCS/revenue codes), pharmacy dispensings (NDC), eligibility spans, and paid amounts link at the person level; a structured outpatient laboratory feed attaches labs to a subset of members.

Why it matters for RWE

Long continuous-enrollment windows support deep baseline lookback; cost fields support economic endpoints without external price benchmarks; MA coverage reaches seniors that FFS-only sources miss. The lab feed enables biomarker-confirmed intermediate outcomes (HbA1c, LDL-C, creatinine) inside a claims architecture.

Operational characteristics

• Enrollment-driven denominators: person-time derives from eligibility spans; distinguish medical-only vs medical+pharmacy enrollment.
• MA dominance among seniors: prior authorization and network effects shift utilization vs FFS Medicare; do not pool casually.
• Cost detail: line-level paid amounts enable cost-of-care and budget-impact analyses.
• Access: licensed flat-file extracts analyzed in researcher environments.

Common pitfalls

• Claims absence is not clinical absence - unmeasured, not normal, where the lab feed does not reach.
• Lab-feed membership is non-random; quantify selection before generalizing biomarker endpoints.
• Employer-plan churn creates artificial turnover unrelated to health status; apply censoring diagnostics.
• Annual refreshes re-number members; lock dataset versions and document cut dates.

Pros, cons, and trade-offs

• vs MarketScan: integrated lab feed and single-payer consistency vs broader multi-employer pooling.
• vs Truveta-style EHR: complete within-plan capture and continuity vs clinical depth from notes/vitals.
• Trade-off: MA-heavy senior reach inherits managed-care artifacts absent from FFS.

When NOT to use

Outcomes needing nuance beyond codes when labs do not cover them; uninsured or non-covered-service questions; sole regulatory evidence without clinical corroboration.

Decision diagram

flowchart LR
  UHG[UnitedHealth ecosystem] --> CM[Clinformatics Data Mart]
  UHG --> LF[Outpatient lab feed] --> LK[Person-level linkage] --> CM
  CM --> A[Cohort build - enrollment-gated]
  A --> O[Safety / effectiveness / cost RWE]

Worked example

Scenario

Estimate 1-year LDL-C goal attainment after initiating evolocumab using claims plus the attached lab feed.

Dataset

Lab-confirmed lipid endpoint among claims-defined initiators.

Steps

Result

42 percent of evaluable evolocumab initiators reached LDL-C under 70 mg/dL vs 21 percent on ezetimibe add-on; lab-feed members skewed urban-commercial, documented as selection limitation.

Trade-offs

Runnable example

\
import pandas as pd

def enroll_gate(elig, pre_days=365):
    idx = elig["index_date"]
    return elig[(elig.enroll_start <= idx - pd.Timedelta(days=pre_days)) &
                (elig.enroll_end   >= idx + pd.Timedelta(days=30))]

def ldl_attainment(labs, cohort, goal=70.0):
    m = labs.merge(cohort[["member_id","index_date"]], on="member_id")
    m = m[(m.panel=="LDLC") &
          (m.result_date >= m.index_date + pd.Timedelta(days=90)) &
          (m.result_date <= m.index_date + pd.Timedelta(days=365))]
    best = m.groupby("member_id").result_value.min().rename("min_ldlc")
    out = cohort.merge(best, left_on="member_id", right_index=True, how="left")
    out["attained"] = (out.min_ldlc < goal).astype("Int64")
    return out

\
library(dplyr)

enroll_gate <- function(elig, pre_days = 365) {
  elig %>% filter(enroll_start <= index_date - days(pre_days),
                  enroll_end   >= index_date + days(30))
}

ldl_attainment <- function(labs, cohort, goal = 70) {
  labs %>% inner_join(cohort %>% select(member_id, index_date), by = "member_id") %>%
    filter(panel == "LDLC", result_date >= index_date + days(90),
           result_date <= index_date + days(365)) %>%
    group_by(member_id) %>% summarise(min_ldlc = min(result_value)) %>%
    right_join(cohort, by = "member_id") %>%
    mutate(attained = as.integer(min_ldlc < goal))
}

\
proc sql;
  create table gated as
  select e.*, i.index_date
  from elig e join idx i on e.member_id = i.member_id
  where e.enroll_start <= i.index_date - 365
    and e.enroll_end   >= i.index_date + 30;

create table attain as
  select g.member_id,
         min(l.result_value) as min_ldlc,
         (calculated min_ldlc < 70) as attained
  from gated g left join labs l
    on l.member_id = g.member_id and l.panel='LDLC'
   and l.result_date between g.index_date+90 and g.index_date+365
  group by g.member_id;
quit;

Citations

1. [1]Dahlen AD, et al. Benchmarking commercial healthcare claims data. medRxiv. 2024.
2. [2]Dahlen AD, et al. Evaluating the generalizability of commercial healthcare claims data. American Journal of Epidemiology. 2025.
3. [3]Optum. Optum products: de-identified data and analytics (Clinformatics).

1. [4]Strom BL. Data validity issues in using claims data. Pharmacoepidemiology and Drug Safety. 2001.