Related Experiment Videos
The accuracy of Scottish Morbidity Record (SMR1) data for identifying hospitalised stroke patients
R J Davenport1, M S Dennis, C P Warlow
1University of Edinburgh, Department of Clinical Neurosciences, Western General Hospital.
Insights
Scottish Morbidity Record (SMR1) data for stroke is reasonably accurate, with 86% sensitivity and 99.9% specificity. Inaccurate clinical coding is a key error source, but SMR1 data is suitable for large-scale stroke audits.
Area of Science:
- Health Informatics
- Epidemiology
- Clinical Auditing
Background:
- Accurate patient data is crucial for effective healthcare management and research.
- The Scottish Morbidity Record (SMR1) is a key source of routinely collected health data in Scotland.
- Validating the accuracy of SMR1 data for specific conditions like stroke is essential.
Purpose of the Study:
- To evaluate the accuracy of SMR1 data for stroke diagnosis.
- To identify reasons for discrepancies in SMR1 stroke coding.
- To compare patient outcomes between SMR1-identified stroke cases and verified stroke cases.
Main Methods:
- Retrospective observational study comparing SMR1 data with a hospital-based stroke register (Lothian Stroke Register).
- Analysis of false positive and false negative SMR1 stroke cases.
- Comparison of 30-day mortality and 56-day discharge home rates between groups.
Main Results:
- SMR1 data demonstrated 86% sensitivity and 99.9% specificity for stroke identification.
- Inaccurate or misleading diagnostic terms from clinicians were primary causes of coding errors.
- No significant differences in key patient outcomes were observed between SMR1-identified and verified stroke groups.
Conclusions:
- Routinely collected SMR1 stroke data is reasonably accurate for clinical auditing purposes.
- Improving clinician education on diagnostic coding can reduce data inaccuracies.
- While generalizability may be limited due to varying coding practices, SMR1 data is a satisfactory tool for large-scale stroke audits.
Objective:
To assess the accuracy of the Scottish Morbidity Record (SMR1) data for stroke by comparing patients with a principal ICD-9 code of stroke on their SMR1 with those registered on our hospital-based stroke register (the Lothian Stroke Register [LSR]). We analysed why false positive and false negative SMR1 cases of stroke arose. We also compared two measures of outcome (death within 30 days, and proportion of patients discharged home within 56 days) in the group with verified stroke with those identified by SMR1 data.
Design:
Retrospective, observational study.
Setting:
A university teaching hospital.
Subjects:
(i) LSR group. We aimed to register all patients admitted to the medical directorate of our hospital with a stroke over a 36 month period. (ii) SMR1 group. All patients with a principal ICD-9 code of stroke on their SMR1 return for the same period.
Results:
566 strokes were registered on the LSR; 84 (15%) of these did not have a principal code of stroke on their SMR1. A further 75 patients not registered on the LSR, but who had a principal code of stroke on their SMR1, were identified; 39 of these had suffered a stroke, 28 had not, and no data were available for eight. Thus, including these missing eight as assumed strokes, the total number of verified strokes was 613; the sensitivity of the SMR1 data was 86%, the specificity 99.9%. Many of the SMR1 false positive and negative cases arose because of inaccurate or misleading diagnostic terms used by medical staff. There were no significant differences for the two outcome measures between the verified group and the SMR1 group.
Conclusions:
Routinely collected SMR1 data for stroke in our hospital was reasonably accurate, but this result may not be widely generalisable as hospitals use different methods of coding. Inadequate data provided by clinicians was an important source of error, and should be correctable with better education. Despite the inaccuracies of the system, based on our results, the SMR1 data are probably a satisfactory way of identifying specific diagnostic groups for large scale audit.