pharmacoepidemiology and drug safety 2015; 24: 38–44 Published online 26 November 2014 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/pds.3726
ORIGINAL REPORT
Evaluating the validity of clinical codes to identify cataract and glaucoma in the UK Clinical Practice Research Datalink Elizabeth M. Kang*, Simone P. Pinheiro, Tarek A. Hammad† and Adel Abou-Ali† US Food and Drug Administration, 10903 New Hampshire Avenue, Silver Spring, MD, 20993, USA
ABSTRACT Purpose The aim of this study is to determine (i) the positive predictive value (PPV) of an algorithm using clinical codes to identify incident glaucoma and cataract events in the Clinical Practice Research Datalink (CPRD) and (ii) the ability to capture the correct timing of these clinical events. Methods A total of 21 339 and 5349 potential cataract and glaucoma cases, respectively, were identified in CPRD between 1 January 1990 and 31 December 2010. Questionnaires were sent to the general practitioners (GP) of 1169 (5.5%) cataract and 1163 (21.7%) glaucoma cases for validation. GPs were asked to verify the diagnosis and the timing of the diagnosis and to provide other supporting information. Results A total of 986 (84.3%) valid cataract questionnaires and 863 (74.2%) glaucoma questionnaires were completed. 92.1% and 92.4% of these used information beyond EMR to verify the diagnosis. Cataract and glaucoma diagnoses were confirmed in the large majority of the cases. The PPV (95% CI) of the cataract and glaucoma Read code algorithm were 92.0% (90.3–93.7%) and 84.1% (81.7–86.6%), respectively. However, timing of diagnosis was incorrect for a substantial proportion of the cases (20.3% and 32.8% of the cataract and glaucoma cases, respectively) among whom 30.4% and 49.2% had discrepancies in diagnosis timing greater than 1 year. Conclusions High PPV suggests that the algorithms based on the clinical Read codes are sufficient to identify the cataract and glaucoma cases in CPRD. However, these codes alone may not be able to accurately identify the timing of the diagnosis of these eye disorders. Copyright © 2014 John Wiley & Sons, Ltd. key words—clinical practice research datalink; read codes; validation of cataract; validation of glaucoma; eye disorders; pharmacoepidemiology Received 24 April 2014; Revised 2 September 2014; Accepted 22 September 2014
INTRODUCTION Electronic medical records (EMR) are widely used in pharmacoepidemiologic studies. However, the validity of the outcome definitions that are solely based on the computerized codes found in the EMR is sometimes uncertain. Relying on sole use of the computerized codes to ascertain the occurrence of the events can lead to misclassification due to rule-out diagnosis, delay in recording, and diagnoses made outside of the data capture. Accurate identification of the outcome is crucial in order to evaluate the drug safety signals. In addition, the ability to determine accurate timing of the events during pre-specified time at risk after the drug exposure is also essential to better understand the
* Correspondence to: E. M. Kang, US Food and Drug Administration, 10903 New Hampshire Avenue, Silver Spring, MD 20993, USA. Email: Elizabeth.Kang@ fda.hhs.gov † AA is currently at Sanofi Pasteur and TAH is at Merck Research Laboratories.
Copyright © 2014 John Wiley & Sons, Ltd.
effect of the drugs. Furthermore, assessing the types of information used by the practitioners to ascertain the diagnoses is especially important to provide confidence that the information beyond what is routinely available in the EMR was considered in the validation assessment. The Clinical Practice Research Datalink (CPRD), formerly known as the General Practice Research Database (GPRD), is a large primary care EMR database in the UK and has been used in many drug safety studies.1,2 CPRD includes detailed longitudinal clinical information including diagnoses, prescriptions, laboratory tests, symptoms, immunizations, hospital referrals, as well as basic sociodemographic characteristics as entered by the general practitioners (GP). In the UK, GP serves as a primary care provider as well as a gatekeeper for other NHS services. Information arising outside of GP practice is likely to be transmitted back to the GP as the GPs coordinate patient’s care and ensure that the patient is managed appropriately. In the UK, regular sight test and eye examinations are conducted by ophthalmic
validation of cataract and glaucoma in cprd
practitioners, either optometrists or ophthalmic medical practitioners. If abnormality or signs of disease is detected by an optometrist, the patient may be referred to an ophthalmologist to conduct further tests, confirm the diagnosis, and plan the treatment.3 All correspondence from the optometrist to the specialists and vice versa is transmitted to the GP. Therefore, medical records and notes maintained by the GP are likely to capture individual patient’s care comprehensively. Researchers have developed algorithms based on the computerized clinical diagnostic codes, such as Read, to define clinical outcomes of interest. Read codes are a coded dictionary of clinical terms that is widely used in the UK National Health Services by both the primary and secondary care sectors. Algorithms based on Read codes that define various diagnoses have been validated previously.4–6 These validation efforts help ensure the quality and credibility of the studies done in CPRD. In this study, we validated cataract and glaucoma cases in support of a cohort study examining antidepressants and risk of eye disorders. Previously, cataract and glaucoma have been studied in CPRD in association with various drugs.7–16 Two previous studies validated ICD codes for cataract diagnoses with high validity (94% and 97%).14,15 Glaucoma was also validated by one study, which confirmed 89% of the cases.17 Our study builds on previous work to validate the Read code-based algorithms used to identify cataract and glaucoma cases in CPRD using survey questionnaires sent to the GPs. Specifically, we determined the positive predictive value (PPV) of the Read codes algorithms to identify incident cataract and glaucoma diagnoses in CPRD, the ability of these algorithms to capture the correct timing of the clinical event, and the sources of the data that the GPs rely on to validate the diagnosis METHODS Case selection The study population was restricted to the patients aged 18 to 80 between 1 January 1990 to 31 December 2010, within the context of a study on the association between antidepressants and eye disorders. We developed algorithms to identify the incident cases of cataract and glaucoma using selected Read codes based on the expert opinion of a consulting ophthalmologist. Diagnoses coded as congenital or traumatic were excluded. Cases likely due to or associated with other disorders or causes were also excluded. Further, to validate only the codes that are more likely to identify definite chronic Copyright © 2014 John Wiley & Sons, Ltd.
39
glaucoma cases, we did not include codes for closedangle glaucoma and ocular hypertension. Surgical cataract codes were not included because the timing of the diagnosis was indeterminable for the patients with these codes. Laterality was not considered; diagnosis codes may indicate disorder in either or both eyes. All Read codes that were used in the validation study are in Appendix 2. To be considered an incident case, a patient must not have cataract or glaucoma diagnoses for at least 1 year prior to incident diagnosis. Additionally, the incident diagnosis date must be at least 1 year after the date the patient registered with the GP practice or the date the practice became up-to standard, whichever came later. Up-to-standard date indicates the point the practice is considered to have continuous and complete recording of patient data. First appearance of a Read code in the EMR for a cataract or glaucoma diagnosis was considered the date of the diagnosis for that particular condition. A total of 21 339 cataract and 5349 glaucoma cases were identified. We randomly selected 1200 (5.6%) cataract and 1200 (22.4%) glaucoma cases for validation. Of these, 31 cataract and 37 glaucoma cases were excluded because their GP no longer participated with CPRD at the time questionnaires were sent to the clinicians for validation. GP questionnaire The questionnaire was designed to ascertain the codeidentified diagnosis of cataract or glaucoma, verify the date of diagnosis, and identify the source of information that GP used to ascertain the diagnosis (e.g. EMR, paper chart, consultation/outpatient letters, hospital/surgical records, and clinical notes). The questionnaire for cataract consisted the following questions: “1) On the basis of your review, do you believe that this patient had cataract?”, “2) please indicate the information sources you used to determine your response to question #1”, “3) if you answered yes in question #1, is the above date [for the cataract diagnosis] correct? if no, provide the correct date of diagnosis”, and “4) was this patient’s cataract diagnosed by an ophthalmologist?”. The GP was also asked to provide photocopies of all relevant hospital summaries, discharge letters, and reports that can verify the diagnosis. The questionnaire for glaucoma consisted of the similar questions adapted to the glaucoma diagnoses. A returned questionnaire was considered invalid if: (i) it was returned blank (i.e. none of the questions were answered) or (ii) each and every question was answered “unknown”. Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
40
e. m. kang et al.
Statistical analysis We calculated response rate as the number of the returned valid questionnaires (i.e. those in which at least one question was answered by the GP) divided by total number of the questionnaires sent. The positive predictive value (PPV) refers to the proportion of the algorithm-identified cases that were confirmed as actual cases by the GP. We calculated PPV as the number of actual/confirmed cases divided by the total number of the algorithm-identified cases for whom a valid questionnaire was received from the patient’s GP. Binominal 95% confidence intervals (CIs) were calculated. We also stratified patients by the transfer out status as recorded in the EMR. Transferred out patients include those who are no longer with the same GP for various reasons, including death. Additionally, we conducted sensitivity analyses excluding borderline glaucoma due to concerns that these may not accurately reflect chronic glaucoma. All analyses were conducted using SAS 9.2 (SAS Institute Inc., Cary, NC, USA).
b
RESULTS Cataract validation A total of 1169 potential cataract cases were selected for validation, and the questionnaires were sent to their GPs (Appendix 1). Of these, 1004 (85.9%) returned the questionnaire, and 18 (1.8%) of the returned questionnaires were considered invalid. Thus, 986 valid questionnaires were received, yielding a response rate of 84.3% (986/1169). In 907 of the 986 valid questionnaires received, the GP confirmed that the patient had indeed been diagnosed with cataract, yielding a PPV (CI) of 92.0% (90.3–93.7%) for the Read code-based algorithm to identify cataracts (Figure 1a). PPVs for the individual Read codes ranged from 67% to 100% (Appendix 2). Many information sources beyond the EMR (i.e. Read codes) were used by the GPs to ascertain the diagnoses. Over half (59%) of the cases were ascertained using consultation letters, followed by clinical notes (42%), electronic medical records (29%), and hospital records (18%) (Figure 2). Paper chart or notes (6%) and other sources (8%) were also used. 41.6% of the GPs used more than one information sources to confirm the diagnosis. Of the confirmed cataract cases, 62.2% answered “Yes” to the question on diagnosis by ophthalmologist. However, our questionnaire was only able to assess whether the diagnosis had been made by an ophthalmologist. Additional assessment of supporting documentation or notes provided by the GPs suggested that a substantial proportion of the GPs who failed to confirm that Copyright © 2014 John Wiley & Sons, Ltd.
Figure 1. a) Proportion of confirmed cataract diagnoses and other validation outcomes according to transfer out status by the time the general practitioner received the study questionnaire. b) Proportion of confirmed glaucoma diagnoses and other validation outcomes according to transfer out status by the time the general practitioner received the study questionnaire
the diagnoses had been made by an ophthalmologist provided some indication that the diagnosis was made by other eye specialists, such as an optician or optometrist, or confirmed by an ophthalmologist Among the confirmed cataract cases, the GPs confirmed the diagnosis date in 704 (77.6%), disagreed with the date in 184 (20.3%), and did not respond or responded unknown in 19 (2.1%) cases. Among the 184 cases in which the GP disagreed with the diagnosis dates, the median difference in timing was ±119 days; 52 (28.3%) cases had difference of less than ±30 days, 56 (30.4%) cases had difference of greater than ±1 year, and 138 (75.0%) of the GP-reported dates occurred before the EMR recorded date (Figure 3). We stratified patients by the transfer out status; 598 cases were still with the same GP and 388 cases had transferred out. Transferred out patients had a lower response rate (80.8% vs 86.8%). The PPV did not differ significantly between the patients who had and had not transferred out of the practice (PPV (CI): 93.6% (91.1–96.0%) and 91.0% (88.7–93.3%), respectively; Figure 1a). Agreement of diagnoses date was also similar across the transfer out status (80.7% Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
validation of cataract and glaucoma in cprd
Figure 2.
41
Sources of information* that the general practitioners (GP) used to confirm the diagnosis of cataract or glaucoma (N = 986 cataract, 863 glaucoma)
Figure 3. Discrepancy in diagnoses dates (in months) between recorded diagnoses dates in CPRD and GP-reported dates among the confirmed cases whose GP provided different date (N = 179 cataract, 235 glaucoma)
vs 75.6% for the transferred out and the non-transferred out patients, respectively). For the non-transferred out patients, 66.1% of the GPs provided additional medical records to support the diagnoses and 95.2% used information/documentation other than the EMR to confirm the diagnoses. For the transferred out patients, only 40.7% of the GPs provided additional records and 87.4% used sources other than the EMR. Likewise, diagnosis by ophthalmologist was lower for the transferred out patients (65.8% vs 56.7%) Glaucoma validation A total of 1163 potential glaucoma cases were selected for validation, and the questionnaires were sent to their GPs (Appendix 1). A total of 875 (75.2%) questionnaires were returned, and 12 (1.4%) of the returned questionnaires were considered invalid. Thus, 863 valid questionnaires were received, yielding a response rate of 74.2% (863/1163). GPs confirmed that the patient Copyright © 2014 John Wiley & Sons, Ltd.
had indeed been diagnosed with glaucoma in 726 of the 863 valid questionnaires received, yielding a PPV (CI) of 84.1% (81.7–86.6%) for the Read code-based algorithm to identify glaucoma (Figure 1b). PPVs for the individual Read codes ranged from 20% to 100% (Appendix 2). Majority of the returned questionnaires used consultation letters (71%), followed by clinical notes (38%) and EMR (34%). Hospital records (17%), paper chart or notes (9%), and other sources (1%) were also used. 46% of the GPs used more than one information source to ascertain the diagnosis. Of the confirmed glaucoma cases, 88.6% answered “Yes” to the question on diagnosis by ophthalmologist. Among the confirmed cases, GP agreed with the diagnosis date identified in the EMR for 469 (64.6%), disagreed for 238 (32.8%), and did not respond or responded unknown for 19 (2.6%) cases. Of 238 confirmed cases with invalid diagnosis dates, the median difference in timing was ±355 days; 57 (24.0%) cases had difference of less than ±30 days, Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
42
e. m. kang et al.
117 (49.2%) cases had difference of greater than ±1 year, and 196 (82.4%) of the GP-reported dates occurred before the EMR recorded date (Figure 3). Patients were stratified by the transfer out status according to the EMR; 616 patients were still with the same GP and 247 patients had transferred out. Patients that remained with the same GP had greater response rate than those who transferred out (76.0% vs 70.2%). The PPV did not differ significantly between the patients who had and had not transferred out of the practice (PPV, 95% CI: 87.9% (83.8–91.9%), 82.6% (79.6–85.6%), respectively; Figure 1b). Agreement of diagnoses date was higher in the transferred out cases (72.8% vs 61.1%). For patients that were still with the same GP, 72.2% provided additional medical records to support the diagnoses and 95.5% used other than the EMR to confirm the diagnoses. For transferred out patients, only 45.8% of the GPs provided additional records and 84.6% used sources other than the EMR. Likewise, diagnosis by ophthalmologist was lower for the transferred out patients (93.5% vs 77.0%). While our primary analyses included borderline glaucoma codes, we conducted additional analyses in which borderline glaucoma codes were excluded. These analyses produced essentially unchanged results. DISCUSSION The results of this validation study suggest that the clinical code based algorithms to identify cataract and glaucoma cases have high PPV (92% and 84%, respectively). The results are consistent with the previous validation studies in CPRD. Derby et. al.15 validated ICD code 374X, mapped from OXMIS code, for diagnosis of cataract in a sample of 225 patients by clinical records (GP records, referral letters from ophthalmic specialists, and hospital discharge summaries), confirming 97% of the cases. In another study, Ruigomez et al.14 validated ICD codes for cataract and cataract surgery (3742, 3749 151–153, 155–157, or 159) in a sample of 29 patients by requesting medical records related to cataracts from GP, confirming 94% of the cases. With regards to glaucoma, Huerta et al.17 mentioned validating cases by sending the questionnaires to the GP, confirming 89% of the cases; definition of glaucoma was not specified in this study. The findings of our study also suggest that majority of the GPs use information beyond what is found in the EMR to validate the diagnosis of cataract and glaucoma and that many diagnoses are actually made by the ophthalmologist. Because eye disorders are likely to be screened for or diagnosed outside of the primary care, it is especially important that the GPs Copyright © 2014 John Wiley & Sons, Ltd.
have available and use additional sources such as the consultation letters from a specialist to validate the cases. If the GPs only examine their computer records, which are the same information that the researchers have available to identify the cases, overestimation of PPV is expected. A small proportion of the GPs used EMR alone and answered “unknown” or “no” to the question on diagnosis by ophthalmologist. One of the reasons mentioned by the GP was the lack of paper records after the patient has left the practice. This is also demonstrated by a slightly higher PPV among the transferred out patients, although the PPV was still acceptable for those who did not transfer out. A large proportion of the diagnosis dates were confirmed by the GP; however, capturing accurate diagnosis timing through the EMR was more challenging than merely determining accuracy of diagnosis. We observed discrepancies between the incident diagnosis date documented in the EMR and the date confirmed by the GP. Discrepancies may be due to, but not limited to, a lag in recording in the EMR or the presence of tentative diagnosis records before the diagnosis is actually confirmed. In addition, as this study relies on the GP’s review of external sources such as consultation letters from ophthalmologist, it is possible that these results over-estimate the discrepancy in the diagnosis dates between the GP reviewed EMR date and the chart date. Consequently, in drug safety studies, it may be difficult to determine whether the exposure to a particular drug actually preceded the outcome and whether the studies using the EMR data are able to capture the relevant risk window accurately. Although a differential discrepancy in timing of diagnosis is not expected to occur across different exposure groups, inaccuracy in dates might be an important issue that would lead to misclassification error and may bias the results toward the null. In our study, the discrepancy in diagnosis timing occurred in 30–40% of cases, and future studies that are sensitive to diagnosis timing should be cautioned. Researchers may consider methods such as lagged exposure time for the main or sensitivity analysis to address the issue. This study has several limitations to consider. First, the asymptomatic nature of cataract and glaucoma and subjectivity in diagnosis may make it difficult for GP to determine the presence and accurate timing of disease. Even when a patient’s diagnosis is confirmed by an ophthalmologist, information transmitted from a specialist to the GP may be misinterpreted or miscoded into EMR at GP practice and result in some misclassification of the diagnoses or timing of the diagnoses. Future studies Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
validation of cataract and glaucoma in cprd
should consider validating with the ophthalmologists to address this concern. Second, specific information on who provided the supporting documentation used by the GPs for case validation was not available for all patients of this study. However, assessment of supporting documentation for sample of patients showed that most supporting documentation including consultation letters were provided by clinicians (or their practices) who conducted the ophthalmologic exam. Third, laterality was not considered in this study. Most Read codes were unspecific to laterality, and confirmation of the disease in either eye by the GP was considered to be a confirmed case. This may alter the PPV or contribute to the discrepancy in the diagnosis dates. Fourth, about 25% of glaucoma and 14% of cataract questionnaires were not returned. Reasons for the unreturned questionnaires were unknown; however, it is unlikely that the patients for whom the GPs returned a questionnaire will differ from those for whom the GP failed to respond to our validation questionnaire. Finally, the GP was slightly more likely to disagree with the diagnosis when the patient was still with the same GP, which may reflect the fact that the GP had more information to refer to. However, PPVs did not differ significantly by the transfer out status. This study built on the previous validation studies and has several strengths. First, this study was first to validate diagnosis dates of cataract and glaucoma in the EMR. Second, we broadened the scope of validation and examined various other aspects such as sources used by the GP for validation and whether the diagnosis was confirmed by an ophthalmologist. Our findings that many GPs used sources beyond EMR reassure that the use of direct mention of Read codes and code-based algorithm are reliable methods to identify cataract and glaucoma cases in CPRD. Third, our study validated about 1000 potential cases for each disorder. To our knowledge, this study is the largest study to date to validate these eye disorders in CPRD. Finally, we validated diagnoses found in CPRD, which is a widely used primary care database for studying various health outcomes in pharmacoepidemiology studies. With GPs as the gatekeepers in UK health system, this primary care EMR provides comprehensive health history of millions of patients. In conclusion, cataract and glaucoma diagnoses were confirmed in a high proportion of patients, and the Read codes for these outcomes are likely to accurately identify true cases. However, accurate diagnosis timing is difficult to determine and studies should consider approaches to account for inaccurate timing. Copyright © 2014 John Wiley & Sons, Ltd.
43
CONFLICTS OF INTEREST The views expressed in this manuscript represent the opinions of the authors and do not necessarily represent the views of the US FDA. All authors were affiliated or employed by the US FDA when the study was completed. KEY POINTS
• •
CPRD Read codes for cataract and glaucoma appear to reliably identify true cases of these conditions. The accuracy of diagnosis dates may not be as reliable. The great majority of GPs use information beyond what is available for researchers in the electronic database to confirm clinical diagnoses to validate cataract or glaucoma diagnoses.
PRIOR POSTINGS AND PRESENTATIONS Preliminary results from this study were presented as a poster in the 28th International Conference on Pharmacoepidemiology and Therapeutic Risk Management in Barcelona, Spain, August 2012. ACKNOWLEDGEMENTS Funding for this study came from the US Food and Drug Administration (FDA). AA received a stipend from the US Department of Energy’s Oak Ridge Institute for Science and Education (ORISE).
REFERENCES 1. García Rodríguez LA, Pérez Gutthann S. Use of the UK General Practice Research Database for pharmacoepidemiology. Br J Clin Pharmacol 1998; 45: 419–425. 2. Clinical Practice Research Datalink. Jan 2014. Bibliography - Research Papers. Available at: http://www.cprd.com/Bibliography/Researchpapers.asp#. Accessed October 3, 2014. 3. Association of Optometrists. Dec 2001. Primary eyecare in the community. Available at: http://www.aop.org.uk/uploads/uploaded_files/primary_eyecare_in_the_ community.pdf. Accessed July 6, 2014. 4. Hammad TA, Margulis AV, Ding Y, Strazzeri MM, Epperly H. Determining the predictive value of Read codes to identify congenital cardiac malformations in the UK Clinical Practice Research Datalink. Pharmacoepidemiol Drug Saf 2013; 22: 1233–1238. doi:10.1002/pds.3511. 5. Hammad TA, McAdams MA, Feight A, Iyasu S, Dal Pan GJ. Determining the predictive value of Read/OXMIS codes to identify incident acute myocardial infarction in the General Practice Research Database. Pharmacoepidemiol Drug Saf 2008; 17: 1197–1201. doi:10.1002/pds.1672. 6. Herrett E, Thomas SL, Schoonen WM, Smeeth L, Hall AJ. Validation and validity of diagnoses in the General Practice Research Database: a systematic review. Br J Clin Pharmacol 2010; 69: 4–14. doi:10.1111/j.1365-2125.2009.03537.x. 7. Miller DP, Watkins SE, Sampson T, Davis KJ. Long-term use of fluticasone propionate/salmeterol fixed-dose combination and incidence of cataracts and glaucoma among chronic obstructive pulmonary disease patients in the UK General Practice Research Database. Int J Chron Obstruct Pulmon Dis 2011; 6: 467–476. doi:10.2147/COPD.S14247.
Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
44
e. m. kang et al.
8. Aina FO, Smeeth L, Hubbard R, Hurt LS, Fletcher AE. Hormone replacement therapy and cataract: a population-based case-control study. Eye Lond Engl 2006; 20: 417–422. doi:10.1038/sj.eye.6701877. 9. Bradbury BD, Lash TL, Kaye JA, Jick SS. Tamoxifen and cataracts: a null association. Breast Cancer Res Treat 2004; 87: 189–196. doi:10.1023/B: BREA.0000041626.76694.85. 10. Smeeth L, Boulis M, Hubbard R, Fletcher AE. A population based case-control study of cataract and inhaled corticosteroids. Br J Ophthalmol 2003; 87: 1247–1251. 11. Smeeth L, Hubbard R, Fletcher AE. Cataract and the use of statins: a case-control study. QJM Mon J Assoc Physicians 2003; 96: 337–343. 12. Schlienger RG, Haefeli WE, Jick H, Meier CR. Risk of cataract in patients treated with statins. Arch Intern Med 2001; 161: 2021–2026. 13. Jick SS, Vasilakis-Scaramozza C, Maier WC. The risk of cataract among users of inhaled steroids. Epidemiol Camb Mass 2001; 12: 229–234. 14. Ruigómez A, García Rodríguez LA, Dev VJ, Arellano F, Raniwala J. Are schizophrenia or antipsychotic drugs a risk factor for cataracts? Epidemiol Camb Mass 2000; 11: 620–623.
Copyright © 2014 John Wiley & Sons, Ltd.
15. Derby L, Maier WC. Risk of cataract among users of intranasal corticosteroids. J Allergy Clin Immunol 2000; 105: 912–916. doi:10.1067/mai.2000.106044. 16. Owen CG, Carey IM, Shah S, et al.. Hypotensive medication, statins, and the risk of glaucoma. Invest Ophthalmol Vis Sci 2010; 51: 3524–3530. doi:10.1167/ iovs.09-4821. 17. Huerta C, García Rodríguez LA, Möller CS, Arellano FM. The risk of obstructive airways disease in a glaucoma population. Pharmacoepidemiol Drug Saf 2001; 10: 157–163. doi:10.1002/pds.567.
SUPPORTING INFORMATION Additional supporting information may be found in the online version of this article at the publisher’s web site.
Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
ORIGINAL REPORT
Evaluating the validity of clinical codes to identify cataract and glaucoma in the UK Clinical Practice Research Datalink Elizabeth M. Kang*, Simone P. Pinheiro, Tarek A. Hammad† and Adel Abou-Ali† US Food and Drug Administration, 10903 New Hampshire Avenue, Silver Spring, MD, 20993, USA
ABSTRACT Purpose The aim of this study is to determine (i) the positive predictive value (PPV) of an algorithm using clinical codes to identify incident glaucoma and cataract events in the Clinical Practice Research Datalink (CPRD) and (ii) the ability to capture the correct timing of these clinical events. Methods A total of 21 339 and 5349 potential cataract and glaucoma cases, respectively, were identified in CPRD between 1 January 1990 and 31 December 2010. Questionnaires were sent to the general practitioners (GP) of 1169 (5.5%) cataract and 1163 (21.7%) glaucoma cases for validation. GPs were asked to verify the diagnosis and the timing of the diagnosis and to provide other supporting information. Results A total of 986 (84.3%) valid cataract questionnaires and 863 (74.2%) glaucoma questionnaires were completed. 92.1% and 92.4% of these used information beyond EMR to verify the diagnosis. Cataract and glaucoma diagnoses were confirmed in the large majority of the cases. The PPV (95% CI) of the cataract and glaucoma Read code algorithm were 92.0% (90.3–93.7%) and 84.1% (81.7–86.6%), respectively. However, timing of diagnosis was incorrect for a substantial proportion of the cases (20.3% and 32.8% of the cataract and glaucoma cases, respectively) among whom 30.4% and 49.2% had discrepancies in diagnosis timing greater than 1 year. Conclusions High PPV suggests that the algorithms based on the clinical Read codes are sufficient to identify the cataract and glaucoma cases in CPRD. However, these codes alone may not be able to accurately identify the timing of the diagnosis of these eye disorders. Copyright © 2014 John Wiley & Sons, Ltd. key words—clinical practice research datalink; read codes; validation of cataract; validation of glaucoma; eye disorders; pharmacoepidemiology Received 24 April 2014; Revised 2 September 2014; Accepted 22 September 2014
INTRODUCTION Electronic medical records (EMR) are widely used in pharmacoepidemiologic studies. However, the validity of the outcome definitions that are solely based on the computerized codes found in the EMR is sometimes uncertain. Relying on sole use of the computerized codes to ascertain the occurrence of the events can lead to misclassification due to rule-out diagnosis, delay in recording, and diagnoses made outside of the data capture. Accurate identification of the outcome is crucial in order to evaluate the drug safety signals. In addition, the ability to determine accurate timing of the events during pre-specified time at risk after the drug exposure is also essential to better understand the
* Correspondence to: E. M. Kang, US Food and Drug Administration, 10903 New Hampshire Avenue, Silver Spring, MD 20993, USA. Email: Elizabeth.Kang@ fda.hhs.gov † AA is currently at Sanofi Pasteur and TAH is at Merck Research Laboratories.
Copyright © 2014 John Wiley & Sons, Ltd.
effect of the drugs. Furthermore, assessing the types of information used by the practitioners to ascertain the diagnoses is especially important to provide confidence that the information beyond what is routinely available in the EMR was considered in the validation assessment. The Clinical Practice Research Datalink (CPRD), formerly known as the General Practice Research Database (GPRD), is a large primary care EMR database in the UK and has been used in many drug safety studies.1,2 CPRD includes detailed longitudinal clinical information including diagnoses, prescriptions, laboratory tests, symptoms, immunizations, hospital referrals, as well as basic sociodemographic characteristics as entered by the general practitioners (GP). In the UK, GP serves as a primary care provider as well as a gatekeeper for other NHS services. Information arising outside of GP practice is likely to be transmitted back to the GP as the GPs coordinate patient’s care and ensure that the patient is managed appropriately. In the UK, regular sight test and eye examinations are conducted by ophthalmic
validation of cataract and glaucoma in cprd
practitioners, either optometrists or ophthalmic medical practitioners. If abnormality or signs of disease is detected by an optometrist, the patient may be referred to an ophthalmologist to conduct further tests, confirm the diagnosis, and plan the treatment.3 All correspondence from the optometrist to the specialists and vice versa is transmitted to the GP. Therefore, medical records and notes maintained by the GP are likely to capture individual patient’s care comprehensively. Researchers have developed algorithms based on the computerized clinical diagnostic codes, such as Read, to define clinical outcomes of interest. Read codes are a coded dictionary of clinical terms that is widely used in the UK National Health Services by both the primary and secondary care sectors. Algorithms based on Read codes that define various diagnoses have been validated previously.4–6 These validation efforts help ensure the quality and credibility of the studies done in CPRD. In this study, we validated cataract and glaucoma cases in support of a cohort study examining antidepressants and risk of eye disorders. Previously, cataract and glaucoma have been studied in CPRD in association with various drugs.7–16 Two previous studies validated ICD codes for cataract diagnoses with high validity (94% and 97%).14,15 Glaucoma was also validated by one study, which confirmed 89% of the cases.17 Our study builds on previous work to validate the Read code-based algorithms used to identify cataract and glaucoma cases in CPRD using survey questionnaires sent to the GPs. Specifically, we determined the positive predictive value (PPV) of the Read codes algorithms to identify incident cataract and glaucoma diagnoses in CPRD, the ability of these algorithms to capture the correct timing of the clinical event, and the sources of the data that the GPs rely on to validate the diagnosis METHODS Case selection The study population was restricted to the patients aged 18 to 80 between 1 January 1990 to 31 December 2010, within the context of a study on the association between antidepressants and eye disorders. We developed algorithms to identify the incident cases of cataract and glaucoma using selected Read codes based on the expert opinion of a consulting ophthalmologist. Diagnoses coded as congenital or traumatic were excluded. Cases likely due to or associated with other disorders or causes were also excluded. Further, to validate only the codes that are more likely to identify definite chronic Copyright © 2014 John Wiley & Sons, Ltd.
39
glaucoma cases, we did not include codes for closedangle glaucoma and ocular hypertension. Surgical cataract codes were not included because the timing of the diagnosis was indeterminable for the patients with these codes. Laterality was not considered; diagnosis codes may indicate disorder in either or both eyes. All Read codes that were used in the validation study are in Appendix 2. To be considered an incident case, a patient must not have cataract or glaucoma diagnoses for at least 1 year prior to incident diagnosis. Additionally, the incident diagnosis date must be at least 1 year after the date the patient registered with the GP practice or the date the practice became up-to standard, whichever came later. Up-to-standard date indicates the point the practice is considered to have continuous and complete recording of patient data. First appearance of a Read code in the EMR for a cataract or glaucoma diagnosis was considered the date of the diagnosis for that particular condition. A total of 21 339 cataract and 5349 glaucoma cases were identified. We randomly selected 1200 (5.6%) cataract and 1200 (22.4%) glaucoma cases for validation. Of these, 31 cataract and 37 glaucoma cases were excluded because their GP no longer participated with CPRD at the time questionnaires were sent to the clinicians for validation. GP questionnaire The questionnaire was designed to ascertain the codeidentified diagnosis of cataract or glaucoma, verify the date of diagnosis, and identify the source of information that GP used to ascertain the diagnosis (e.g. EMR, paper chart, consultation/outpatient letters, hospital/surgical records, and clinical notes). The questionnaire for cataract consisted the following questions: “1) On the basis of your review, do you believe that this patient had cataract?”, “2) please indicate the information sources you used to determine your response to question #1”, “3) if you answered yes in question #1, is the above date [for the cataract diagnosis] correct? if no, provide the correct date of diagnosis”, and “4) was this patient’s cataract diagnosed by an ophthalmologist?”. The GP was also asked to provide photocopies of all relevant hospital summaries, discharge letters, and reports that can verify the diagnosis. The questionnaire for glaucoma consisted of the similar questions adapted to the glaucoma diagnoses. A returned questionnaire was considered invalid if: (i) it was returned blank (i.e. none of the questions were answered) or (ii) each and every question was answered “unknown”. Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
40
e. m. kang et al.
Statistical analysis We calculated response rate as the number of the returned valid questionnaires (i.e. those in which at least one question was answered by the GP) divided by total number of the questionnaires sent. The positive predictive value (PPV) refers to the proportion of the algorithm-identified cases that were confirmed as actual cases by the GP. We calculated PPV as the number of actual/confirmed cases divided by the total number of the algorithm-identified cases for whom a valid questionnaire was received from the patient’s GP. Binominal 95% confidence intervals (CIs) were calculated. We also stratified patients by the transfer out status as recorded in the EMR. Transferred out patients include those who are no longer with the same GP for various reasons, including death. Additionally, we conducted sensitivity analyses excluding borderline glaucoma due to concerns that these may not accurately reflect chronic glaucoma. All analyses were conducted using SAS 9.2 (SAS Institute Inc., Cary, NC, USA).
b
RESULTS Cataract validation A total of 1169 potential cataract cases were selected for validation, and the questionnaires were sent to their GPs (Appendix 1). Of these, 1004 (85.9%) returned the questionnaire, and 18 (1.8%) of the returned questionnaires were considered invalid. Thus, 986 valid questionnaires were received, yielding a response rate of 84.3% (986/1169). In 907 of the 986 valid questionnaires received, the GP confirmed that the patient had indeed been diagnosed with cataract, yielding a PPV (CI) of 92.0% (90.3–93.7%) for the Read code-based algorithm to identify cataracts (Figure 1a). PPVs for the individual Read codes ranged from 67% to 100% (Appendix 2). Many information sources beyond the EMR (i.e. Read codes) were used by the GPs to ascertain the diagnoses. Over half (59%) of the cases were ascertained using consultation letters, followed by clinical notes (42%), electronic medical records (29%), and hospital records (18%) (Figure 2). Paper chart or notes (6%) and other sources (8%) were also used. 41.6% of the GPs used more than one information sources to confirm the diagnosis. Of the confirmed cataract cases, 62.2% answered “Yes” to the question on diagnosis by ophthalmologist. However, our questionnaire was only able to assess whether the diagnosis had been made by an ophthalmologist. Additional assessment of supporting documentation or notes provided by the GPs suggested that a substantial proportion of the GPs who failed to confirm that Copyright © 2014 John Wiley & Sons, Ltd.
Figure 1. a) Proportion of confirmed cataract diagnoses and other validation outcomes according to transfer out status by the time the general practitioner received the study questionnaire. b) Proportion of confirmed glaucoma diagnoses and other validation outcomes according to transfer out status by the time the general practitioner received the study questionnaire
the diagnoses had been made by an ophthalmologist provided some indication that the diagnosis was made by other eye specialists, such as an optician or optometrist, or confirmed by an ophthalmologist Among the confirmed cataract cases, the GPs confirmed the diagnosis date in 704 (77.6%), disagreed with the date in 184 (20.3%), and did not respond or responded unknown in 19 (2.1%) cases. Among the 184 cases in which the GP disagreed with the diagnosis dates, the median difference in timing was ±119 days; 52 (28.3%) cases had difference of less than ±30 days, 56 (30.4%) cases had difference of greater than ±1 year, and 138 (75.0%) of the GP-reported dates occurred before the EMR recorded date (Figure 3). We stratified patients by the transfer out status; 598 cases were still with the same GP and 388 cases had transferred out. Transferred out patients had a lower response rate (80.8% vs 86.8%). The PPV did not differ significantly between the patients who had and had not transferred out of the practice (PPV (CI): 93.6% (91.1–96.0%) and 91.0% (88.7–93.3%), respectively; Figure 1a). Agreement of diagnoses date was also similar across the transfer out status (80.7% Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
validation of cataract and glaucoma in cprd
Figure 2.
41
Sources of information* that the general practitioners (GP) used to confirm the diagnosis of cataract or glaucoma (N = 986 cataract, 863 glaucoma)
Figure 3. Discrepancy in diagnoses dates (in months) between recorded diagnoses dates in CPRD and GP-reported dates among the confirmed cases whose GP provided different date (N = 179 cataract, 235 glaucoma)
vs 75.6% for the transferred out and the non-transferred out patients, respectively). For the non-transferred out patients, 66.1% of the GPs provided additional medical records to support the diagnoses and 95.2% used information/documentation other than the EMR to confirm the diagnoses. For the transferred out patients, only 40.7% of the GPs provided additional records and 87.4% used sources other than the EMR. Likewise, diagnosis by ophthalmologist was lower for the transferred out patients (65.8% vs 56.7%) Glaucoma validation A total of 1163 potential glaucoma cases were selected for validation, and the questionnaires were sent to their GPs (Appendix 1). A total of 875 (75.2%) questionnaires were returned, and 12 (1.4%) of the returned questionnaires were considered invalid. Thus, 863 valid questionnaires were received, yielding a response rate of 74.2% (863/1163). GPs confirmed that the patient Copyright © 2014 John Wiley & Sons, Ltd.
had indeed been diagnosed with glaucoma in 726 of the 863 valid questionnaires received, yielding a PPV (CI) of 84.1% (81.7–86.6%) for the Read code-based algorithm to identify glaucoma (Figure 1b). PPVs for the individual Read codes ranged from 20% to 100% (Appendix 2). Majority of the returned questionnaires used consultation letters (71%), followed by clinical notes (38%) and EMR (34%). Hospital records (17%), paper chart or notes (9%), and other sources (1%) were also used. 46% of the GPs used more than one information source to ascertain the diagnosis. Of the confirmed glaucoma cases, 88.6% answered “Yes” to the question on diagnosis by ophthalmologist. Among the confirmed cases, GP agreed with the diagnosis date identified in the EMR for 469 (64.6%), disagreed for 238 (32.8%), and did not respond or responded unknown for 19 (2.6%) cases. Of 238 confirmed cases with invalid diagnosis dates, the median difference in timing was ±355 days; 57 (24.0%) cases had difference of less than ±30 days, Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
42
e. m. kang et al.
117 (49.2%) cases had difference of greater than ±1 year, and 196 (82.4%) of the GP-reported dates occurred before the EMR recorded date (Figure 3). Patients were stratified by the transfer out status according to the EMR; 616 patients were still with the same GP and 247 patients had transferred out. Patients that remained with the same GP had greater response rate than those who transferred out (76.0% vs 70.2%). The PPV did not differ significantly between the patients who had and had not transferred out of the practice (PPV, 95% CI: 87.9% (83.8–91.9%), 82.6% (79.6–85.6%), respectively; Figure 1b). Agreement of diagnoses date was higher in the transferred out cases (72.8% vs 61.1%). For patients that were still with the same GP, 72.2% provided additional medical records to support the diagnoses and 95.5% used other than the EMR to confirm the diagnoses. For transferred out patients, only 45.8% of the GPs provided additional records and 84.6% used sources other than the EMR. Likewise, diagnosis by ophthalmologist was lower for the transferred out patients (93.5% vs 77.0%). While our primary analyses included borderline glaucoma codes, we conducted additional analyses in which borderline glaucoma codes were excluded. These analyses produced essentially unchanged results. DISCUSSION The results of this validation study suggest that the clinical code based algorithms to identify cataract and glaucoma cases have high PPV (92% and 84%, respectively). The results are consistent with the previous validation studies in CPRD. Derby et. al.15 validated ICD code 374X, mapped from OXMIS code, for diagnosis of cataract in a sample of 225 patients by clinical records (GP records, referral letters from ophthalmic specialists, and hospital discharge summaries), confirming 97% of the cases. In another study, Ruigomez et al.14 validated ICD codes for cataract and cataract surgery (3742, 3749 151–153, 155–157, or 159) in a sample of 29 patients by requesting medical records related to cataracts from GP, confirming 94% of the cases. With regards to glaucoma, Huerta et al.17 mentioned validating cases by sending the questionnaires to the GP, confirming 89% of the cases; definition of glaucoma was not specified in this study. The findings of our study also suggest that majority of the GPs use information beyond what is found in the EMR to validate the diagnosis of cataract and glaucoma and that many diagnoses are actually made by the ophthalmologist. Because eye disorders are likely to be screened for or diagnosed outside of the primary care, it is especially important that the GPs Copyright © 2014 John Wiley & Sons, Ltd.
have available and use additional sources such as the consultation letters from a specialist to validate the cases. If the GPs only examine their computer records, which are the same information that the researchers have available to identify the cases, overestimation of PPV is expected. A small proportion of the GPs used EMR alone and answered “unknown” or “no” to the question on diagnosis by ophthalmologist. One of the reasons mentioned by the GP was the lack of paper records after the patient has left the practice. This is also demonstrated by a slightly higher PPV among the transferred out patients, although the PPV was still acceptable for those who did not transfer out. A large proportion of the diagnosis dates were confirmed by the GP; however, capturing accurate diagnosis timing through the EMR was more challenging than merely determining accuracy of diagnosis. We observed discrepancies between the incident diagnosis date documented in the EMR and the date confirmed by the GP. Discrepancies may be due to, but not limited to, a lag in recording in the EMR or the presence of tentative diagnosis records before the diagnosis is actually confirmed. In addition, as this study relies on the GP’s review of external sources such as consultation letters from ophthalmologist, it is possible that these results over-estimate the discrepancy in the diagnosis dates between the GP reviewed EMR date and the chart date. Consequently, in drug safety studies, it may be difficult to determine whether the exposure to a particular drug actually preceded the outcome and whether the studies using the EMR data are able to capture the relevant risk window accurately. Although a differential discrepancy in timing of diagnosis is not expected to occur across different exposure groups, inaccuracy in dates might be an important issue that would lead to misclassification error and may bias the results toward the null. In our study, the discrepancy in diagnosis timing occurred in 30–40% of cases, and future studies that are sensitive to diagnosis timing should be cautioned. Researchers may consider methods such as lagged exposure time for the main or sensitivity analysis to address the issue. This study has several limitations to consider. First, the asymptomatic nature of cataract and glaucoma and subjectivity in diagnosis may make it difficult for GP to determine the presence and accurate timing of disease. Even when a patient’s diagnosis is confirmed by an ophthalmologist, information transmitted from a specialist to the GP may be misinterpreted or miscoded into EMR at GP practice and result in some misclassification of the diagnoses or timing of the diagnoses. Future studies Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
validation of cataract and glaucoma in cprd
should consider validating with the ophthalmologists to address this concern. Second, specific information on who provided the supporting documentation used by the GPs for case validation was not available for all patients of this study. However, assessment of supporting documentation for sample of patients showed that most supporting documentation including consultation letters were provided by clinicians (or their practices) who conducted the ophthalmologic exam. Third, laterality was not considered in this study. Most Read codes were unspecific to laterality, and confirmation of the disease in either eye by the GP was considered to be a confirmed case. This may alter the PPV or contribute to the discrepancy in the diagnosis dates. Fourth, about 25% of glaucoma and 14% of cataract questionnaires were not returned. Reasons for the unreturned questionnaires were unknown; however, it is unlikely that the patients for whom the GPs returned a questionnaire will differ from those for whom the GP failed to respond to our validation questionnaire. Finally, the GP was slightly more likely to disagree with the diagnosis when the patient was still with the same GP, which may reflect the fact that the GP had more information to refer to. However, PPVs did not differ significantly by the transfer out status. This study built on the previous validation studies and has several strengths. First, this study was first to validate diagnosis dates of cataract and glaucoma in the EMR. Second, we broadened the scope of validation and examined various other aspects such as sources used by the GP for validation and whether the diagnosis was confirmed by an ophthalmologist. Our findings that many GPs used sources beyond EMR reassure that the use of direct mention of Read codes and code-based algorithm are reliable methods to identify cataract and glaucoma cases in CPRD. Third, our study validated about 1000 potential cases for each disorder. To our knowledge, this study is the largest study to date to validate these eye disorders in CPRD. Finally, we validated diagnoses found in CPRD, which is a widely used primary care database for studying various health outcomes in pharmacoepidemiology studies. With GPs as the gatekeepers in UK health system, this primary care EMR provides comprehensive health history of millions of patients. In conclusion, cataract and glaucoma diagnoses were confirmed in a high proportion of patients, and the Read codes for these outcomes are likely to accurately identify true cases. However, accurate diagnosis timing is difficult to determine and studies should consider approaches to account for inaccurate timing. Copyright © 2014 John Wiley & Sons, Ltd.
43
CONFLICTS OF INTEREST The views expressed in this manuscript represent the opinions of the authors and do not necessarily represent the views of the US FDA. All authors were affiliated or employed by the US FDA when the study was completed. KEY POINTS
• •
CPRD Read codes for cataract and glaucoma appear to reliably identify true cases of these conditions. The accuracy of diagnosis dates may not be as reliable. The great majority of GPs use information beyond what is available for researchers in the electronic database to confirm clinical diagnoses to validate cataract or glaucoma diagnoses.
PRIOR POSTINGS AND PRESENTATIONS Preliminary results from this study were presented as a poster in the 28th International Conference on Pharmacoepidemiology and Therapeutic Risk Management in Barcelona, Spain, August 2012. ACKNOWLEDGEMENTS Funding for this study came from the US Food and Drug Administration (FDA). AA received a stipend from the US Department of Energy’s Oak Ridge Institute for Science and Education (ORISE).
REFERENCES 1. García Rodríguez LA, Pérez Gutthann S. Use of the UK General Practice Research Database for pharmacoepidemiology. Br J Clin Pharmacol 1998; 45: 419–425. 2. Clinical Practice Research Datalink. Jan 2014. Bibliography - Research Papers. Available at: http://www.cprd.com/Bibliography/Researchpapers.asp#. Accessed October 3, 2014. 3. Association of Optometrists. Dec 2001. Primary eyecare in the community. Available at: http://www.aop.org.uk/uploads/uploaded_files/primary_eyecare_in_the_ community.pdf. Accessed July 6, 2014. 4. Hammad TA, Margulis AV, Ding Y, Strazzeri MM, Epperly H. Determining the predictive value of Read codes to identify congenital cardiac malformations in the UK Clinical Practice Research Datalink. Pharmacoepidemiol Drug Saf 2013; 22: 1233–1238. doi:10.1002/pds.3511. 5. Hammad TA, McAdams MA, Feight A, Iyasu S, Dal Pan GJ. Determining the predictive value of Read/OXMIS codes to identify incident acute myocardial infarction in the General Practice Research Database. Pharmacoepidemiol Drug Saf 2008; 17: 1197–1201. doi:10.1002/pds.1672. 6. Herrett E, Thomas SL, Schoonen WM, Smeeth L, Hall AJ. Validation and validity of diagnoses in the General Practice Research Database: a systematic review. Br J Clin Pharmacol 2010; 69: 4–14. doi:10.1111/j.1365-2125.2009.03537.x. 7. Miller DP, Watkins SE, Sampson T, Davis KJ. Long-term use of fluticasone propionate/salmeterol fixed-dose combination and incidence of cataracts and glaucoma among chronic obstructive pulmonary disease patients in the UK General Practice Research Database. Int J Chron Obstruct Pulmon Dis 2011; 6: 467–476. doi:10.2147/COPD.S14247.
Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
44
e. m. kang et al.
8. Aina FO, Smeeth L, Hubbard R, Hurt LS, Fletcher AE. Hormone replacement therapy and cataract: a population-based case-control study. Eye Lond Engl 2006; 20: 417–422. doi:10.1038/sj.eye.6701877. 9. Bradbury BD, Lash TL, Kaye JA, Jick SS. Tamoxifen and cataracts: a null association. Breast Cancer Res Treat 2004; 87: 189–196. doi:10.1023/B: BREA.0000041626.76694.85. 10. Smeeth L, Boulis M, Hubbard R, Fletcher AE. A population based case-control study of cataract and inhaled corticosteroids. Br J Ophthalmol 2003; 87: 1247–1251. 11. Smeeth L, Hubbard R, Fletcher AE. Cataract and the use of statins: a case-control study. QJM Mon J Assoc Physicians 2003; 96: 337–343. 12. Schlienger RG, Haefeli WE, Jick H, Meier CR. Risk of cataract in patients treated with statins. Arch Intern Med 2001; 161: 2021–2026. 13. Jick SS, Vasilakis-Scaramozza C, Maier WC. The risk of cataract among users of inhaled steroids. Epidemiol Camb Mass 2001; 12: 229–234. 14. Ruigómez A, García Rodríguez LA, Dev VJ, Arellano F, Raniwala J. Are schizophrenia or antipsychotic drugs a risk factor for cataracts? Epidemiol Camb Mass 2000; 11: 620–623.
Copyright © 2014 John Wiley & Sons, Ltd.
15. Derby L, Maier WC. Risk of cataract among users of intranasal corticosteroids. J Allergy Clin Immunol 2000; 105: 912–916. doi:10.1067/mai.2000.106044. 16. Owen CG, Carey IM, Shah S, et al.. Hypotensive medication, statins, and the risk of glaucoma. Invest Ophthalmol Vis Sci 2010; 51: 3524–3530. doi:10.1167/ iovs.09-4821. 17. Huerta C, García Rodríguez LA, Möller CS, Arellano FM. The risk of obstructive airways disease in a glaucoma population. Pharmacoepidemiol Drug Saf 2001; 10: 157–163. doi:10.1002/pds.567.
SUPPORTING INFORMATION Additional supporting information may be found in the online version of this article at the publisher’s web site.
Pharmacoepidemiology and Drug Safety, 2015; 24: 38–44 DOI: 10.1002/pds
