The Record Is Not the Whole Patient
Coverage changes several outcomes at once. Spending can mislabel need. Diagnosis counts can rise when the system looks harder.
Editorial mapping
UNH / UnitedHealth Group
The company association indexes this file within the five-pair editorial system. It does not indicate sponsorship, endorsement, or a scientific conclusion.
Clinical question
How do insurance access, algorithmic targets, and regional diagnostic intensity shape the outcomes recorded by a health system?
Structured source reading
This page does not claim recorded human clinical review.
What gets confused when labels get lazy
Administrative data are built by care, access, payment, and measurement processes. Treating every field as a direct reading of health can turn a convenient proxy into a confident error.
How this file was read
This dossier compares a randomized natural experiment in insurance access, an empirical audit of a population-health algorithm, and an observational natural experiment using moves between regions. It avoids corrected numerical values that could not be rechecked in the publisher notice.
Read across the three files
The coverage experiment found different patterns across clinical measures, service use, depression screening, detection and management, and financial strain; no single outcome summarized the policy effect. The algorithm study showed how predicting spending instead of illness burden could reproduce unequal access within a risk score. The regional analysis found that recorded diagnoses and risk scores changed with local diagnostic intensity, indicating that claims reflect both illness and how actively a system searches. Together, the studies make one disciplined point: coverage, cost, and diagnosis codes are informative, but none is a complete proxy for health need or care quality.
Three studies. Labels attached.
The Oregon Experiment — Effects of Medicaid on Clinical Outcomes
Baicker K, Taubman SL, Allen HL, et al.; Oregon Health Study Group. N Engl J Med. 2013;368:1713–1722.
- Study design
- Randomized natural experiment using lottery selection for the opportunity to apply for Medicaid; effects were estimated from lottery assignment over approximately two years.
- Question
- How did access to insurance affect measured clinical outcomes, depression screening, service use, and financial strain among low-income adults in Oregon?
- Finding
- The study did not demonstrate significant improvement in several measured physical-health outcomes, while service use, diabetes detection and management, depression-screening results, and financial strain showed different patterns. One endpoint did not summarize the entire coverage effect.
- Limits
- Lottery selection did not insure every selected person. The findings are tied to a specific low-income adult population and follow-up period; nonsignificant physical-health results do not establish that insurance has no health benefit.
DOI 10.1056/NEJMsa1212321
PMID 23635051
Direct source record
Oregon Health Study Group. The Oregon experiment--effects of Medicaid on clinical outcomes. 2013.
- DOI
- 10.1056/NEJMsa1212321
- PMID
- 23635051
- Authoritative source
- PubMed abstract and bibliographic record
- Accessed
Dissecting racial bias in an algorithm used to manage the health of populations
Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Science. 2019;366:447–453.
- Study design
- Retrospective empirical evaluation of a deployed population-health risk algorithm, comparing its risk score with measured illness burden and health spending.
- Question
- Can predicting healthcare cost as a proxy for health need introduce bias into decisions about who receives additional care?
- Finding
- At the same algorithmic risk score, Black patients carried greater illness burden. Using spending as the target allowed unequal access to care to enter the risk score.
- Limits
- The analysis concerns a specific algorithm and dataset; the same magnitude cannot be assumed for every model or current implementation. Changing the target does not prove that every inequity disappears, and this was not a treatment-effect trial.
DOI 10.1126/science.aax2342
PMID 31649194
Direct source record
Obermeyer Z and colleagues. Dissecting racial bias in an algorithm used to manage the health of populations. 2019.
- DOI
- 10.1126/science.aax2342
- PMID
- 31649194
- Authoritative source
- PubMed abstract and bibliographic record
- Accessed
Regional Variations in Diagnostic Practices
Song Y, Skinner J, Bynum J, Sutherland J, Wennberg JE, Fisher ES. N Engl J Med. 2010;363:45–53.
- Study design
- Observational natural experiment using Medicare claims to compare diagnosis, testing, and risk-score patterns before and after beneficiaries moved between regions.
- Question
- Are diagnosis counts used in risk adjustment affected by regional differences in how intensively clinicians investigate and record disease?
- Finding
- People moving to higher-intensity regions had larger increases in recorded diagnoses and risk scores. Claims can therefore reflect diagnostic practice as well as underlying illness burden.
- Limits
- Moves were not randomized, and the data came from a historical fee-for-service Medicare population. The study does not establish that diagnoses were wrong or that more investigation is always wasteful. Corrected table values are deliberately not reproduced here.
DOI 10.1056/NEJMsa0910881
PMID 20463332
Direct source records
Song Y and colleagues. Regional Variations in Diagnostic Practices. 2010.
- DOI
- 10.1056/NEJMsa0910881
- PMID
- 20463332
- Authoritative source
- PubMed Central full text
- Accessed
New England Journal of Medicine. Correction: Regional Variations in Diagnostic Practices. 2010.
- DOI
- 10.1056/NEJMx100034
- Authoritative source
- Publisher correction DOI; direct access was blocked during the latest check
- Accessed
Where the evidence stops
The evidence comes from specific historical populations, payment arrangements, and systems. Lottery selection did not guarantee enrollment, algorithm findings cannot be assigned to every model, and moving regions was not randomized. These records do not support an insurer-quality ranking, investment conclusion, or present-day causal estimate for another health system.
What was available at the desk
The Oregon Experiment — Effects of Medicaid on Clinical Outcomes
PubMed abstract and bibliographic record
Dissecting racial bias in an algorithm used to manage the health of populations
PubMed abstract and bibliographic record
Regional Variations in Diagnostic Practices
PubMed Central full text
Regional Variations in Diagnostic Practices
Publisher correction DOI; direct access was blocked during the latest check
Corrections travel with the finding
- Regional Variations in Diagnostic Practices
A publisher correction exists at DOI 10.1056/NEJMx100034 and concerns risk-score table values and a result sentence. Because the correction page could not be reopened, no disputed numerical value or percentage is transferred into this dossier.
