Counting hospital activity: spells or episodes?

Paul Aylin, Susan Williams, Alex Bottle, Brian Jarman

References

w1 Mckenna CJ, Forfar JC. Was it a heart attack? BMJ 2002;324:377–8.

w2 Department of Health.Implementing payment by results. Technical guidance 2004/5. 2004. www.dh.gov.uk/assetRoot/04/08/49/35/04084935.pdf (accessed 9 Nov 2004).

w3 Dixon, J. Payment by results – new financial flows in the NHS. BMJ 2004;328:969-70

[posted as supplied by author]

Methods

The Dr Foster case note is based on analysis of Hospital Episode Statistics (HES). HES data is routinely collected within the health service for administrative purposes and not specifically for clinical audit. There may be issues around coverage, completeness and accuracy that need to be considered when interpreting the results. HES data are submitted by all NHS hospital trusts in England and capture all admissions, both inpatients and day cases, and also include deaths occurring in hospital, recording details such as the patient’s age and diagnosis and any operations they had. The data are in the form of finished consultant episodes (the continuous period during which the patient is under the care of one consultant), which need to be linked into "spells" or admissions. About 14% of inpatient spells comprise more than one episode, and the patient’s method and date of discharge need to be derived from the final episode in the spell.

Data fields can contain missing or invalid values and spells may be incomplete. Episodes are assumed to be duplicates if they have the same combination of provider, date of birth, sex, postcode, date of admission and episode number (PROCODE, DOB, SEX, HOMEADD, EPISTART, EPIEND, EPIORDER). Duplicates are excluded at this stage.

Some spells have the same date of admission (ADMIDATE) but different dates of discharge (DISDATE). This is not valid unless the patient was discharged and readmitted on the same day, and the spell with the earliest ADMIDATE was arbitrarily taken to be the valid one. Episodes relating to the invalid spell are excluded at this stage.

When episodes are linked together to form spells, for years before 2000/1, records are assumed to belong to the same person if they match on date of birth, sex and postcode (DOB, SEX, HOMEADD) as the NHS number is not complete or accurate enough. Episodes with invalid DOB (recorded by HES as 15th Oct 1582) or SEX (i.e. not ‘1’ or ‘2’) are excluded at this stage. From 2000/1 we have been given HESID, derived by the Department of Health, and have been using this as a patient identifier. Only ages within the ranges 1-120 and 7001-7007 (special values to indicate age in months for children under 1 year) are considered valid.

Also excluded at this stage are unfinished episodes (EPISTAT=1), unknown/invalid method of admission (ADMIMETH <> 11,12,13,21,22,23,24,28,81). Episode linkage is not performed on day cases, which are discarded for projects involving only inpatients.

Remaining episodes are sorted by provider, date of birth, sex, postcode, date of admission, date of discharge and episode number (PROCODE, DOB, SEX, HOMEADD, ADMIDATE, DISDATE, EPIORDER). Episodes are not required to be in strict sequence, only in chronological order. For example, if the first one had EPIORDER=01, the second one had EPIORDER=03 and the last one of the same spell had EPIORDER=99, then the three episodes are treated just the same as if they were numbered 01, 02 and 03 (as most multi-episode spells are). The dataset is then split into the first and last episodes of each spell (which are often the same, as most spells comprise only one episode). Diagnosis and procedure variables are taken from the first episode (DIAG1-DIAG7, OPER1-OPER4, date of primary procedure OP_DTE_1).

Mention of ICD10 codes I21 (Acute MI) and I22 (Subsequent MI) in the primary diagnosis of each episode was used to select episodes for this analysis.