| Verified against primary record | |
| The basic form | A ratio of the observed number of deaths to the expected number of deaths for a provider[1] |
|---|---|
| Variables in one national model | Age, gender, admission method, year index, Charlson Comorbidity Index and diagnosis grouping[1] |
| What its publisher says it is not | Not a measure of quality of care[2] |
| Records read | A model specification, two statistical releases, two methods documents and two comparison studies, 9 October 2026 |
| Independently reported | |
| Sensitivity to coding | 70.3 per cent of hospitals changed position in a mortality league table after a validation audit[3] |
| Bands apply only to the rows beneath them. No risk-adjustment model comparing a destination hospital with a home hospital was located. | |
Risk adjustment, also called case-mix adjustment, is the statistical correction applied before one hospital’s results are set beside another’s. It exists because hospitals treat different patients, and it is the reason a raw comparison of outcomes between a destination hospital and a home hospital carries almost no information.
How it works
The commonest form is a ratio. England’s summary hospital-level mortality indicator is defined as a ratio of the observed number of deaths to the expected number of deaths for a trust, where the expected figure comes from a model fitted on a patient case-mix of age, gender, admission method, year index, Charlson Comorbidity Index and diagnosis grouping.[1] A value above one means more deaths than the model predicted for that mix of patients.
The reason for doing this is stated plainly by the body that runs England’s outcome measures programme: comparing unadjusted average scores between providers can be misleading, because the patient profiles that one provider treats may be different to the patient profile at another provider.[4] Adjustment is an attempt to answer what the difference would have been had both hospitals treated the same patients.
What the publishers say their own indicators cannot do
The caveats attached to these indicators by the organisations that produce them are stronger than the uses the indicators are commonly put to. Of the English mortality indicator its publisher states that it is not a measure of quality of care, that a higher than expected number of deaths should not immediately be interpreted as indicating poor performance, and that the difference between observed and expected deaths cannot be interpreted as the number of avoidable or excess deaths for the trust.[2]
One omission is documented explicitly. The methodology does not make any adjustment for patients recorded as receiving palliative care, because there is considerable variation between trusts in the way that palliative care is recorded.[5] Rather than adjust for a variable it cannot trust, the publisher leaves it out and issues a separate contextual indicator alongside. That is a candid treatment of an unresolvable problem, and it illustrates the general one: a risk model can only adjust for what is recorded consistently.
Coding variation moves the answer
Because the expected figure is built from coded diagnoses, how thoroughly a hospital codes changes its apparent performance. A French study quantified this by auditing a sample: of 70,402 discharge abstracts, 715 were recoded against full medical records. Hospital comorbidity-level positive predictive values ranged from 64.4 per cent to 96.4 per cent, and after correction 70.3 per cent of hospitals changed position in the mortality league table. Using the Charlson index, 61.5 per cent changed position. The authors conclude that variations in administrative data coding can bias mortality comparisons and budget allocation across hospitals.[3]
That finding is the practical ceiling on this method. A league table in which seven hospitals in ten move when a one per cent sample is recoded is measuring documentation practice as much as care.
Small numbers
A second limit bites hardest on exactly the kind of provider a medical traveller might consider: one performing a modest annual volume of a particular procedure. American quality-indicator methodology applies smoothing, which brings rates toward the mean and does this more so for outliers such as rural hospitals, with the weight closer to zero for small providers, and retains a covariate only where there are at least 30 cases with the outcome of interest.[6] England’s outcome programme states that at lower sample sizes the models are unstable, and reports providers performing fewer than 30 revision procedures in raw form only.[4]
Smoothing is honest, and it has a consequence worth naming: a small hospital’s adjusted figure is pulled toward the average, so it will rarely look either excellent or poor, whatever it is actually doing.
Across borders
One project has produced risk-adjusted outcome rates for an international sample of hospitals, pooling 6,737,211 inpatient records including 214,622 in-hospital deaths for 2005 to 2010, with in-hospital mortality, unplanned readmission and prolonged stay adjusted by logistic regression. Its authors record that although diagnostic coding depth varied appreciably by country, comorbidity weights were broadly comparable, and caution that intercountry differences in outcomes may result from differences in the quality of care but may equally reflect practice patterns.[7] It was built for mutual learning among volunteering hospitals, not for patients choosing between countries.
A five-country feasibility study reached a more negative conclusion about routine comparison, finding that different indicators are collected in each country and that different definitions of the same indicators are used, and stating that further work is urgently needed to agree the way forward and that until then patients will not be able to make informed choices.[8]
Searches for this entry did not surface a validated risk-adjustment model designed to compare a named destination hospital with a named home-country hospital for an individual patient’s procedure. That is a search result rather than proof that none exists, but it means any provider claim of superior adjusted outcomes relative to a patient’s home system currently rests on no shared model.
See also
- Treatment success rates in medical tourism, case mix as one of four reasons rates differ
- Clinical outcomes in medical tourism, what would be adjusted if it were collected
- Selection bias in medical tourism research, the problem adjustment cannot fix
- Medical tourism rankings, the same instability in composite league tables
References
- Health and Social Care Information Centre, Clinical Indicators Team. Summary Hospital-level Mortality Indicator specification. Version 1.19, 10 June 2015. Verified against primary record: model specification opened and read. Retrieved 9 October 2026.
- NHS England Digital. Summary Hospital-level Mortality Indicator, deaths associated with hospitalisation. Statistical release, August 2026. Verified against primary record: publisher’s own caveats read in the release. Retrieved 9 October 2026.
- Haviari S, Chollet F, Polazzi S, Payet C, Beauveil A, Colin C, Duclos A. Effect of data validation audit on hospital mortality ranking and pay for performance. BMJ Quality and Safety, 2019;28(6):459-467, DOI 10.1136/bmjqs-2018-008039. Independently reported: peer-reviewed study. Retrieved 9 October 2026.
- NHS England. Patient Reported Outcome Measures in England, update to reporting and case-mix adjusting hip and knee procedure data. October 2013, with the programme guide of August 2018. Verified against primary record: case-mix methodology opened and read. Retrieved 9 October 2026.
- NHS England Digital. Summary Hospital-level Mortality Indicator, palliative care coding contextual indicators. August 2026, covering discharges April 2025 to March 2026. Verified against primary record: contextual indicator note opened and read. Retrieved 9 October 2026.
- Agency for Healthcare Research and Quality. Quality Indicator empirical methods. Revised November 2014. Verified against primary record: smoothing and covariate rules opened and read. Retrieved 9 October 2026.
- Bottle A, Middleton S, Kalkman CJ, Livingston EH, Aylin P. Global Comparators Project, international comparison of hospital outcomes using administrative data. Health Services Research, 2013. Independently reported: peer-reviewed study, abstract and programme record read. Retrieved 9 October 2026.
- Burnett S, Renz A, Wiig S, Fernandes A, Weggelaar AM, Calltorp J, Andersen J, Robert G, Vincent CA, Fulop NJ. Prospects for comparing European hospitals in terms of quality and safety, lessons from a comparative study in five countries. International Journal for Quality in Health Care, 2013;25(1):1-7. Independently reported: peer-reviewed study, abstract and metadata read. Retrieved 9 October 2026.
Sourcing note: the model specification, the two statistical releases, the two methods documents and the two comparison studies were opened and read on 9 October 2026. Two of the eight sources were read at abstract and metadata level only, and are marked as such in the list. The volume number for the validation-audit study was taken from the journal’s own article path; the pages and digital object identifier are as printed. No model comparing a destination hospital with a home hospital was located, and that is reported as the result of searching rather than as proof that none exists.