Differentiating Incident from Recurrent Stroke Using Administrative Data: The Impact of Varying Lengths of Look-Back Periods on the Risk of Misclassification

Abstract

Background/Aims: Administrative data are widely used to monitor epidemiological trends in stroke and outcomes; yet there is scant empirical guidance on how to best differentiate incident from recurrent stroke. Methods: We identified all hospital admissions in New South Wales, Australia, with a principal stroke diagnosis from July 1, 2013 to June 30, 2014, linked to 12 years of previous admissions. We calculated the proportion of cases identified with a prior stroke to determine the number of years of look-back required to minimise misclassification of incident and recurrent strokes. Results: Using the maximum available look-back period of 12 years, 1,171 out of 8,364 eligible stroke cases (14.0%) had a stroke history. A 1-year look-back period identified only 25.1% of these patients and 1 in 10 stroke cases were misclassified as incident. With a 10-year clearance period, less than 1 in 100 stroke cases were misclassified as incident. The risk of misclassification was lower in patients younger than 65 years and in those with haemorrhagic stroke. Conclusion: Hospital administrative data sets linked to prior admissions can be used to distinguish recurrent from incident stroke. The risk of misclassifying recurrent stroke cases as incident events is negligible with a look-back period of 10 years.

Measuring stroke outcomes for 74 501 patients using linked administrative data: System‐wide estimates and validation of ‘home‐time’ as a surrogate measure of functional status

“…'Home-time' may therefore underestimate disability owing to cognitive deficits, a shortcoming shared with traditional stroke outcome measures, such as the modified Rankin Scale score. 41 (15)(16)(17)(18)(19)(20) .006 25 (20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32) .12…”

“…The study was part of a larger research project, the H 2 O Study, which aimed to report stroke epidemiology and outcomes in New South Wales (NSW), Australia's most populous state (population ~7.99 million in 2019). 18,19 The project utilised four sources of routinely collected health data linked to allow individual patient-level data analysis as described below.…”

“…stroke admission and previous admissions; (b) mortality data, obtained from the NSW Registry of Births, Deaths and Marriages, to determine fact of death in the 90 days after stroke; (c) a measure of stroke severity obtained from the vital signs survey recorded by first-responder paramedics, obtained from prehospital ambulance data and (d) functional impairment, recorded upon admission to rehabilitation, obtained from the NSW Sub-acute and Non-Acute Patient Data Collection. As previously described, 18,19 deterministic and probabilistic linkages were performed by the Centre for Health Record Linkage (CHeReL) using gold-standard privacy preserving protocols and ensuring that the linkage error rate was no greater than 5 per 1000. Patient identifiers, such as names, addresses, date of birth and dates of health service delivery (eg, dates of admission) facilitated linkage.…”

“…Subtype was defined in this manner for 2417 or 2.5% of patients; otherwise, we adhered to recommendations to code these patients as ischaemic stroke (ie, I64) 21 or intracerebral haemorrhage (ie, I62.9). 22 As previously reported, 18,19 patients were ineligible if resident outside NSW or if patients had concomitantly recorded major head trauma (ICD-10AM S04, S06, S07, S08, S17, S18, S01. were also excluded as missing hospitalisation data precluded accurate calculation of 'home-time'.…”

“…No 42 (42)(43) 21 (20)(21)(22) 31 (29)(30)(31)(32) Peripheral vascular disease Yes 39 (37)(38)(39)(40) <.001 18 (16)(17)(18)(19)(20) .01 30 (22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35)(36)(37)(38)(39)(40)(41) .88…”