Comparison of comorbidities of stroke collected in administrative data, surveys, clinical trials and cohort studies

Abstract

Background Administrative data are used extensively for research purposes, but there remains limited information on the quality of these data for identifying comorbidities related to stroke. Objective To compare the prevalence of comorbidities of stroke identified using International Classification Diseases, Australian Modification (ICD-10-AM) or Anatomical Therapeutic Chemical codes, with those from (i) self-reported data and (ii) published studies. Method The cohort included patients with stroke or transient ischaemic attack admitted to hospitals (2012–2016; Victoria and Queensland) in the Australian Stroke Clinical Registry (N = 26,111). Data were linked with hospital and pharmaceutical datasets to ascertain comorbidities using published algorithms. The sensitivity, specificity, and positive predictive value of these comorbidities were compared with survey responses from 623 patients (reference standard). An indirect comparison was also performed with clinical data from published stroke studies. Results The sensitivity of hospital ICD-10-AM data was poor for most comorbidities, except for diabetes (93.0%). Specificity was excellent for all comorbidities (87–96%), except for hypertension (70.5%). Compared to published stroke studies (3 clinical trials and 1 incidence study), the prevalence of diabetes and atrial fibrillation in our cohort was similar using ICD-10-AM codes, but lower for dyslipidaemia and anxiety/depression. Whereas in the pharmaceutical dispensing data, the sensitivity was excellent for dyslipidaemia (94%) and modest for anxiety/depression (77%). In the pharmaceutical data, specificity was modest for hypertension (78%) and anxiety or depression (76%), but specificity was poor for dyslipidaemia (19%) and heart disease (46%). Conclusion Variation was observed in the reporting of comorbidities of stroke in administrative data, and consideration of multiple sources of data may be necessary for research. Further work is needed to improve coding and clinical documentation for reporting of comorbidities in administrative data.

Factors Associated With Transition From Community to Permanent Residential Aged Care Following Stroke: A Linked Registry Data Study

“…Comorbidities with low sensitivity in the coded hospital data (eg, hypertension) were supplemented using pharmaceutical dispensing records of relevant medications based on Anatomical Therapeutic Chemical codes. 14 Comorbidities were also combined to derive a weighted Charlson comorbidity index score. 15,16 The regularity of primary care physician visits was calculated for the 2 years before stroke using a validated method for persistence.…”

Clinical documentation integrity: Its role in health data integrity, patient safety and quality outcomes and its impact on clinical coding and health information management

“…For CCs and HIMs , the integrity of clinical documentation is critical for accurate clinical coding. Kilkenny et al (2022) reported variation in the reporting of comorbidities of stroke in administrative data suggesting the need to improve clinical documentation. Duke et al (2022) examined coding-based definitions for sepsis and identified significant differences in sepsis classification in administrative data (e.g.…”

“…Hosseini et al (2022) argued that data quality enhancement requires collaboration between physicians, nurses, managers and developers of software, suggesting a concept mapping approach to meet this objective. Kilkenny et al (2023) recommended ongoing education for CCs to achieve high-quality coding and more accurate data. Campbell and Giadresco (2020) reviewed 39 articles on computer-assisted clinical coding (CAC) and found it improved both accuracy and quality of data.…”

“…Hay et al (2020) reported that only 55% of laboratory-confirmed episodes of Staphylococcus aureus bacteraemia were reflected in the coded data, suggesting that inaccuracies were most likely due to documentation issues. Kilkenny et al (2022, 2023) examined the importance of accurate coded diagnostic data for epidemiological research into stroke and suggested that given the variation in reporting of comorbidities of stroke in administrative data, multiple sources of data may be necessary for research. While these datasets are critical, in many cases their users have little or no understanding of how the data are collected (Hemsley and Debono, 2022).…”

“…Recommendations indicated that clinician education could be extended beyond initial training, be ongoing and encompass clinical classifications and coding standards (Hosseini et al, 2022). Education for other stakeholders included targeted training for CCs (Doktorchik et al, 2020; Kilkenny et al, 2023), together with CDI audits to monitor, raise awareness and provide feedback (Alonso et al, 2020; McLachlan et al, 2023). The importance of building strong professional relationships and communication is imperative for collaborative working in all CDI activities, according to Doktorchik et al (2020), Hosseini et al (2022) and Pine et al (2023).…”