Results from the validation study have been published in Shah et al.17 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.
Introduction
The first phase of this study was to carry out an internal validation of shoulder dislocation coding in routinely collected data from the CPRD to confirm the feasibility of using the CPRD to study shoulder dislocations for the main analyses. It also sought to identify which risk factors were relevant to shoulder dislocation and were readily available in the CPRD for use in the main analyses.
Objectives of Chapter 2
- Identify patients in the CPRD aged 16–35 years who were diagnosed with a traumatic shoulder dislocation between 1 April 1997 and 31 March 2015 in England.
- Develop a GP questionnaire using a validation algorithm and with input from GPs.
- Take a random sample of patients from those identified in the CPRD and use the CPRD GP questionnaire service to send the questionnaire to the respective patient practices for completion.
Methods
Data source
Population-based primary care data from the CPRD were used to identify a cohort of patients diagnosed with a traumatic shoulder dislocation (aged 16–35 years) in the UK from 1 April 1997 to 31 March 2015. The CPRD covers 11.3 million people from 674 UK general practices and provides a representative coverage of around 6.9% of the UK population, which is broadly representative of the population in terms of age, sex and ethnicity.18 Patient and practice data are anonymised, but patient-level data are available on age, sex, geographic region, body mass index (BMI), smoking status (i.e. current smoker, ex-smoker, non-smoker) and drinking status (i.e. current drinker, ex-drinker, non-drinker). The CPRD data were linked to data from the Index of Multiple Deprivation (IMD) 200419 for English patients and the Charlson Comorbidity Index (CCI) score was calculated using predefined Read codes.
Participants
The Read codes were used to identify patients from the CPRD with a shoulder dislocation. These codes had been established a priori through consensus by specialist shoulder surgeons with clinical experience and experts in epidemiology research (see Appendix 1). To ensure that ‘primary’ or ‘first-time’ shoulder dislocations were captured, patients were required to have no recorded shoulder dislocations in their CPRD clinical data for 2 years prior to first entry of a shoulder dislocation Read code. The 2-year washout period was defined using the date that the general practice was classified as ‘up to standard’ and the date that the patient was first registered at the general practice. This first entry of a shoulder dislocation Read code was defined as the primary dislocation.
The patients had to be registered at ‘active’ CPRD practices. An ‘active’ practice was defined as a practice that had contributed to the CPRD database in the previous 6 months. No general practices were classified as active in the East Midlands, and so it was not possible to include patients from this region. Following the identification of the shoulder dislocation cohort from within the CPRD data set, a predefined set of patient exclusion criteria was applied to facilitate the validation (Table 1).
| Exclusions | n (%) |
|---|---|
| Total number of CPRD shoulder dislocation patients received | 63,324 (100) |
| Unacceptable patients (i.e. CPRD flags that data quality for a patient is insufficient for medical research) | 806 (1) |
| Unacceptable dates (i.e. impossible to find a shoulder dislocation code between the CPRD minimum and maximum acceptable dates as defined by data management standard operating procedures for clinical research)18 | 34,710 (55) |
| Shoulder dislocation date prior to 1 April 1997 | 2507 (4) |
| Shoulder dislocation date after 31 March 2015 | 823 (1) |
| < 2-year minimum washout period (i.e. washout period defined using the date the general practice was classified as ‘up to standard’ and the date the patient first registered at the practice) | 3694 (6) |
| Aged < 16 years | 825 (1) |
| Aged > 35 years | 12,125 (19) |
| Non-resident of England patients | 1788 (3) |
| Patients remaining in cohort | 6046 (10) |
General practitioner questionnaire design and implementation
The GP questionnaire was designed with the assistance of GPs and based on a developed validation algorithm (see Appendices 2 and 3) and a random sample of 172 patients was selected from a list of the 6046 eligible patients. CPRD personnel then sent the questionnaire to each patient’s general practice for a clinician to complete by comparing the records on their CPRD computer system with the patient’s clinical records. The GPs assessed the use of shoulder dislocation codes for traumatic dislocation, confirmation of first-time shoulder dislocation, subsequent codes used for further events, physiotherapy referral codes and confirmation that physiotherapy took place.
Four written reminders were sent every 2 weeks to the general practices. Data from the returned questionnaires were double-entered into a data set by a statistician and a project manager. Any queries were resolved by an academic orthopaedic shoulder surgeon. In the instance that two questionnaires were received for the same patient with differing answers (on three occasions two questionnaires were sent back for one patient, i.e. three patients and six questionnaires; differing responses were only received for questions 6 and 7), clarification was sought from the general practice via CPRD personnel (clarification was received for one patient).
The following validation criteria were defined a priori to reflect that the coding of shoulder dislocations in the CPRD was of a sufficiently high quality to proceed with the main analyses planned in work package 2:
- a positive predictive value of ≥ 75% accuracy for shoulder dislocation coding in the CPRD
- a positive predictive value of ≥ 75% accuracy in coding ‘primary’ or ‘first-time’ traumatic shoulder dislocation within the CPRD.
Results
Cohort
An initial cohort of 63,324 patients with codes for shoulder dislocation was identified from the CPRD database. A database manager and statistician assessed the cohort against clear predefined and important exclusion criteria (see Table 1). Unacceptable patients were defined as those whose records had not met quality standards and had been flagged by the CPRD as ‘unacceptable’. Unacceptable dates were defined as when it was impossible to find a shoulder dislocation code between the CPRD minimum and maximum acceptable dates, as defined by data management standard operating procedures for clinical research.18 During this process, the majority of patients were excluded, either because they had a shoulder dislocation diagnosis outside the study time period (55%) or because they were outside the study age limits (16–35 years) (20%). The final cohort included 6046 patients aged 16–35 years who were diagnosed with a shoulder dislocation between 1 April 1997 and 31 March 2015 in England.
Internal validation
Of the 172 patients whose GP received a copy of the validation questionnaire, a response for 95 (55%) patients was received. For two patients, their GPs confirmed that they had transferred out of the practice and that no further information was available for them on the CPRD system.
Table 2 presents demographic characteristics for the following patient groups:
- the cohort of 6046 patients from the CPRD aged 16–35 years and diagnosed with a shoulder dislocation between 1 April 1997 and 31 March 2015 in England
- the 172 patients randomly selected to have their GPs receive a questionnaire
- the 97 patients for whom completed questionnaires were returned by their general practice
- the 75 patients for whom questionnaires were not returned by their general practice.
| Demographic characteristic | Patient group | |||
|---|---|---|---|---|
| Whole cohort | All GP questionnaires | Responders | Non-responders | |
| Cohort size, (n) | 6046 | 172 | 97 | 75 |
| Sex, n (%) | ||||
| Male | 4991 (83) | 137 (80) | 81 (84) | 56 (75) |
| Female | 1055 (17) | 35 (20) | 16 (16) | 19 (25) |
| Median age (years) (IQR) | 24 (20–34) | 24 (20–29) | 24 (20–29) | 24 (19–29) |
| Median BMI (kg/m2) (IQR) | 24 (22–27) | 24 (21–27) | 25 (22–28) | 23 (21–26) |
| Median CCI score (IQR) | 0 (0–0) | 0 (0–0) | 0 (0–0) | 0 (0–0) |
| Region, n (%) | ||||
| East Midlands | 263 (4) | 0 (0) | 0 (0) | 0 (0) |
| East of England | 673 (11) | 21 (12) | 17 (18) | 4 (5) |
| London | 695 (12) | 23 (13) | 11 (11) | 12 (16) |
| North East | 133 (2) | 4 (2) | 3 (3) | 1 (1) |
| North West | 951 (16) | 29 (17) | 13 (13) | 16 (21) |
| South Central | 965 (16) | 32 (19) | 17 (18) | 15 (20) |
| South East Coast | 702 (12) | 25 (15) | 12 (12) | 13 (17) |
| South West | 743 (12) | 17 (10) | 17 (18) | 0 (0) |
| West Midlands | 667 (11) | 15 (9) | 7 (7) | 8 (11) |
| Yorkshire and the Humber | 254 (4) | 6 (3) | 0 (0) | 6 (8) |
| IMD 2004 quintile, n (%) | ||||
| 1 (affluent) | 1279 (21) | 53 (31) | 28 (29) | 26 (35) |
| 2 | 1077 (18) | 35 (20) | 20 (21) | 15 (20) |
| 3 | 958 (16) | 24 (14) | 16 (16) | 8 (11) |
| 4 | 876 (14) | 38 (22) | 19 (20) | 18 (24) |
| 5 (deprived) | 624 (10) | 22 (13) | 14 (14) | 8 (11) |
| Missing | 1232 (20) | 0 (0) | 0 (0) | 0 (0) |
All four groups were similar with respect to demographic characteristics, including age, BMI and CCI score. The highest response rate (100%) was received from general practices in the South West of England, but otherwise the proportion of responses received reflected the regional distribution of patients included in the cohort. Data on the IMD 200419 were obtained and linked after the selection of the 172 records to be validated. There were no missing data on deprivation, as all practices sampled were ‘active practices’. A higher proportion of patients had been sampled from category 1 (affluent) and category 4 (somewhat deprived) than in the initial cohort of 6046 patients, but otherwise response rates were similar from all deprivation groups.
The distribution of CPRD Read codes used by GPs to code shoulder dislocations is given in Table 3. Codes S41..00, S41z.00 and 14G5.00 for dislocation of shoulder accounted for 82% of all shoulder dislocation coding in the data. Recurrent shoulder dislocation codes only identified another 10% of patients, indicating that the 2-year washout period was a successful approach to identifying primary or first-time shoulder dislocations. Of the seven patients who had a recurrent shoulder dislocation code and for whom a GP questionnaire response was obtained, four were confirmed as having had a primary shoulder dislocation and one was confirmed as not having had a shoulder dislocation at all.
| CPRD description (Read code) | Patient group (%) | |||
|---|---|---|---|---|
| Whole cohort | All GP questionnaires | Responders | Non-responders | |
| Total number of patients | 6046 | 172 | 97 | 75 |
| CPRD description (Read code) | ||||
| Dislocation or subluxation of shoulder (S41..00) | 55 | 55 | 52 | 60 |
| Dislocation of shoulder NOSa (S41z.00) | 10 | 10 | 9 | 11 |
| H/O:a dislocated shoulder (14G5.00) | 17 | 19 | 20 | 17 |
| Closed reduction of dislocation of shoulder (7K6G300) | 3 | 3 | 5 | 1 |
| Closed traumatic dislocation of shoulder (S410.00) | 2 | 1 | 1 | 0 |
| Recurrent dislocation of shoulder, anterior (N083A00) | 6 | 6 | 5 | 7 |
| Anterior dislocation of shoulder (S410111) | 2 | 1 | 2 | 0 |
| Recurrent joint dislocation, of shoulder region (N083100) | 2 | 0 | 0 | 0 |
| Recurrent subluxation of shoulder, anterior (N083C00) | 2 | 2 | 2 | 3 |
| Closed traumatic dislocation shoulder joint, anterior (subcoracoid) (S410100) | < 1 | 1 | 0 | 1 |
| Closed traumatic dislocation shoulder joint, unspecified (S410000) | < 1 | 1 | 2 | 0 |
| Closed traumatic subluxation, shoulder (S412.00) | < 1 | 1 | 2 | 0 |
Shoulder dislocation was confirmed as having been coded correctly in 89% (95% CI 83% to 95%) of all patients (Table 4). The remaining 11% (10 patients) had been miscoded and the patient had suffered other shoulder trauma or injuries, such as strains or dislocations of the acromioclavicular joint, as confirmed by their GP. Of all patients, a first-time or primary shoulder dislocation was confirmed in 76% (95% CI 67% to 85%) of cases. Subsequent dislocations occurring up to 2 years after the primary dislocation were recorded in the CPRD for 32% of patients. From the GP responses, an additional 11% of patients experienced a re-dislocation during this time that was not recorded in the CPRD.
| Validation of shoulder dislocations | n (%) |
|---|---|
| GP confirmation of shoulder dislocation | 85 (89) |
| Patients who had a confirmed ‘primary’ shoulder dislocation | 72 (76) |
| Patients who had a further dislocation within 2 years of the primary dislocationa | 27 (32) |
| Confirmation that this was a further dislocation episode and not a review of the problem | 21 (78) |
| Patients with further dislocations that have not been noted in the CPRD | 9 (11) |
| Patients who have CPRD Read codes for physiotherapy in the 2 years following the first dislocation code | 24 (28) |
| It is clear that this physiotherapy code indicates that the patient received physiotherapy for their shoulder | 15 (63) |
| Patients who did not have a CPRD Read code for physiotherapy, but for whom documentation exists confirming that they received physiotherapy for their shoulder | 17 (21) |
Twenty-eight per cent of patients had been coded as having received physiotherapy within the CPRD, and GPs confirmed that physiotherapy had been given to 63% of these. However, a further 17 patients had received physiotherapy for their shoulder dislocation that was not recorded within the CPRD. Thus, 41% of patients known to be receiving physiotherapy were not recorded within the CPRD.
Conclusion
This validation exercise, carried out, to our knowledge, for the first time for this condition in the CPRD, has demonstrated that the CPRD is an acceptable data set to identify and study shoulder dislocation patients. The validity of GP coding of shoulder dislocations within the CPRD in a subset of patients proved very high, at 89%. Of all patients, 76% were confirmed to have primary shoulder dislocations. All of the CPRD Read codes used to identify shoulder dislocation patients were useful for identifying patients who had a primary shoulder dislocation, including the three codes that are specific to re-dislocations (i.e. N083A00, N083100 and N083C00). There was a small amount of under-reporting of subsequent shoulder dislocations. Physiotherapy treatment coding was of a poorer quality given that it is under-reported, at 41%, and, as such, the effectiveness of physiotherapy cannot be evaluated using the CPRD in any subsequent analyses of shoulder dislocations. Although not all general practices responded to the questionnaire, those that did and those that did not respond to the questionnaire survey were similar by deprivation level, geography and other demographic characteristics.
The strength of the CPRD is that it is a large, population-based primary care cohort that is representative of the UK general population. The positive internal validation result achieved on the correct coding of shoulder dislocations in the CPRD now provides the opportunity to use these codes and study definitions in the main study analysis.