Clinical Epidemiology Dovepress open access to scientific and medical research Open Access Full Text Article ORIGINAL RESEARCH How well can electronic health records from primary care identify Alzheimer’s disease cases? Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 This article was published in the following Dove Press journal: Clinical Epidemiology Anna Ponjoan 1–3 Background: Electronic health records (EHR) from primary care are emerging in Josep Garre-Olmo 3 Alzheimer’s disease (AD) research, but their accuracy is a concern. We aimed to Jordi Blanch 1 validate AD diagnoses from primary care using additional information provided by general Ester Fages 1,4 practitioners (GPs), and a register of dementias. Lia Alves-Cabratosa 1 Patients and methods: This retrospective observational study obtained data from the System for the Development of Research in Primary Care (SIDIAP). Three algorithms Ruth Martí-Lluch 1–3 For personal use only. combined International Statistical Classification of Diseases (ICD-10) and Anatomical Marc Comas-Cufí 1 Therapeutic Chemical codes to identify AD cases in SIDIAP. GPs evaluated dementia Dídac Parramon 1,4 diagnoses by means of an online survey. We linked data from the Register of Dementias María García-Gil 1 of Girona and from SIDIAP. We estimated the positive predictive value (PPV) and sensitivity Rafel Ramos 1,5 and provided results stratified by age, sex and severity. 1 Vascular Health Research Group (ISV- Results: Using survey data from the GPs, PPV of AD diagnosis was 89.8% (95% CI: Girona), Jordi Gol Institute for Primary 84.7–94.9). Using the dataset linkage, PPV was 74.8 (95% CI: 73.1–76.4) for algorithm A1 Care Research (IDIAPJGol), Barcelona, Catalonia, Spain; 2Universitat Autònoma (AD diagnoses), and 72.3 (95% CI: 70.7–73.9) for algorithm A3 (diagnosed or treated de Barcelona, Bellaterra, Catalonia, Spain; patients without previous conditions); sensitivity was 71.4 (95% CI: 69.6–73.0) and 83.3 3 Girona Biomedical Research Institute (95% CI: 81.8–84.6) for algorithms A1 (AD diagnoses) and A3, respectively. Stratified (IDIBGI), Girona, Catalonia, Spain; 4 Primary Care Services, Catalan Health results did not differ by age, but PPV and sensitivity estimates decreased amongst men and Institute (ICS), Girona, Catalonia, Spain; severe patients, respectively. 5 Department of Medical Sciences, School Conclusions: PPV estimates differed depending on the gold standard. The development of of Medicine, Campus Salut, University of Girona, Girona, Catalonia, Spain algorithms integrating diagnoses and treatment of dementia improved the AD case ascertain- ment. PPV and sensitivity estimates were high and indicated that AD codes recorded in a large primary care database were sufficiently accurate for research purposes. Keywords: dementia, family physician, survey, algorithm, data accuracy, real-world data, validation, electronic medical records Introduction Alzheimer’s disease (AD) entails a heavy burden for patients, their families and public health systems. Its prevalence and economic costs are forecasted to increase dramatically in coming decades due to worldwide population aging.1,2 Electronic health records (EHR) might be useful to update the AD epidemiology, especially in primary care settings. Paradoxically, although general practitioners (GPs) are the gatekeepers of health care services and thus play a pivotal role in the recognition of AD, primary care EHR have scarcely been applied in dementia studies, and even Correspondence: Rafel Ramos less frequently in AD studies.3,4 Unitat de Suport a la Recerca – IDIAPJGol, c/Maluquer Salvador, 11 Baixos, Girona A concern about using EHR is the accuracy of diagnoses, as reported in research 17002, Catalonia, Spain Tel +3 497 248 7968 on dementia diagnoses in primary care,5 but recent studies report high positive Email rramos.girona.ics@gencat.cat predictive value (PPV), indicating that dementia codes in primary care databases submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 509–518 509 DovePress © 2019 Ponjoan et al. This work is published and licensed by Dove Medical Press Limited. The full terms of this license are available at https://www.dovepress.com/terms. php and incorporate the Creative Commons Attribution – Non Commercial (unported, v3.0) License (http://creativecommons.org/licenses/by-nc/3.0/). By accessing the http://doi.org/10.2147/CLEP.S206770 work you hereby accept the Terms. Non-commercial uses of the work are permitted without any further permission from Dove Medical Press Limited, provided the work is properly attributed. For permission for commercial use of this work, please see paragraphs 4.2 and 5 of our Terms (https://www.dovepress.com/terms.php). Powered by TCPDF (www.tcpdf.org) Ponjoan et al Dovepress were sufficiently accurate for research use.3,6 Nonetheless, cases (Table 1). We considered treated patients as cases literature about the accuracy of EHR in AD is mainly because in Catalonia the prescription of anti-dementia focused on hospital records7–9 and data from primary drugs can be requested by the GP but requires approval care is scant.4,10 from a geriatrician, a psychiatrist, or a neurologist. We hypothesized that EHR from primary care would be We followed two approaches. First, we asked the GPs a valid tool to study AD, since dementia diagnoses are to complete an online survey to provide additional infor- accurate6 and AD is the most prevalent dementia subtype. mation about dementia diagnoses.6 Our previously pub- The present study sought to validate AD diagnoses recorded lished validation study focused on diagnoses of overall Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 in a primary care database by comparing AD from primary dementia recorded in SIDIAP; here, we provide a sub- care with information from an online survey of GPs and analysis restricted to AD diagnoses (International with data from a registry of dementia diagnoses. Statistical Classification of Diseases, 10th revision (ICD- 10): F00, G30). Second, we linked SIDIAP data with a dementia-specific registry for the province of Girona. Materials and methods We conducted a retrospective observational study, approved by the Clinical Research Ethics Committee of the Primary Survey sent to GPs Care Research Institute IDIAP Jordi Gol (IDIAPJGol). We Survey questions, design and administration are described followed the STARDS 2015 guidelines11 and the RECORD elsewhere.6 Briefly, GPs from the IDIAPJGol Agency of statement.12 Clinical Research Management in Primary Care (AGICAP) For personal use only. We used data from the Information System for network20 were invited to participate. Those who agreed to Research in Primary Care (SIDIAP database), which con- participate were sent the online survey in January 2018 and tains anonymized longitudinal medical records related to given 1 month to evaluate a consecutive series of their patients demographics including deprivation index,13 symptoms, with a dementia diagnosis recorded in the EHR.6 The survey diagnoses, laboratory tests, prescriptions, and pharmacy asked about the current basis for a dementia diagnosis, the dispensing from about 6 million patients (>80% of the fulfillment of DSM-IV or ICD-10 diagnosis criteria, and the Catalan population) attended in the primary care settings dementia subtype (Figure S1). Patients were defined as AD of the Catalan Health Service.14,15 Accuracy of SIDIAP cases if they were identified as a true dementia case (diagnosed data have been analyzed for several conditions16–18 includ- by a specialist; or based on cognitive and functional tests or ing overall dementia,6 but not yet for AD. a clinical opinion) and the GP confirmed the dementia subtype Using SIDIAP data, we identified AD cases using algo- as AD. Such confirmation of dementia subtype was based on rithms that combined EHR, a method previously applied to the fulfillment of the DSM-IVor the ICD10 diagnostic criteria. identify dementia cases.10,19 We followed Imfeld et al 2013 The inclusion of both sets of diagnostic criteria aimed to to define three algorithms that combine information about facilitate the identification of true cases, because some physi- diagnoses and pharmacological treatment to identify AD cians might be more familiarized with one of them. Table 1 Algorithms to identify Alzheimer's disease (AD) cases in electronic health records from primary care, using codes from the international statistical classification of diseases and related health problems, version 10 (ICD-10) and from the Anatomical Therapeutic Chemical (ATC) Algorithm Definition of AD case A1 Diagnosed patients: have an ICD10 code for AD (F00 or G30). A2 Diagnosed or treated patients: have a code for AD (ICD10: F00 or G30) or for prescription or billing of anti-dementia drugs (ATC: N06DA, N06DX01). A3 Diagnosed or treated patients without previous conditions: have a code for AD (ICD10: F00 or G30) or for prescription or billing of anti-dementia drugs (ATC: N06DA, N06DX01). Treated patients were included if they had no code of dementia diagnosis or had a code of unspecified dementia (F03), and were excluded if they had a code for: a specific subtype of dementia such as Lewy bodies dementia, vascular or frontotemporal dementia (ICD10: F01, F02); Parkinson (ICD10: G20-G22); anti-Parkinson drugs (ATC: N04); or cerebrovascular disease (ICD10:I60- I69, G45, G46) within two years prior to AD diagnosis. submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 510 DovePress Powered by TCPDF (www.tcpdf.org) Dovepress Ponjoan et al Linkage between SIDIAP and the registry In ReDeGi, AD cases were identified according to ICD-10 codes, excluding mixed dementia cases. In of dementias SIDIAP, we identified AD cases using the above- The Registry of Dementias of Girona (ReDeGi) contains mentioned algorithms combining EHR (Table 1). demographic and clinical data of all the incident cases of dementia diagnosed in the seven hospitals of the public health care system within the province of Girona.21 The Statistical analysis We used percentages to describe categorical variables and ReDeGi uses standardized criteria for case definition, and mean (SD) for continuous variables. Using data from the follows the Center for Disease Control and Prevention Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 survey of GPs, we calculated the PPV including the num- guidelines on surveillance systems.22 The ReDeGi ber of confirmed diagnoses of AD in the numerator, includes the Clinical Dementia Rating score,23 a measure against all the diagnoses of AD evaluated by GP, in the of disease severity (mild, moderate and severe). A recent denominator. We replicated the PPV estimate study found high adherence to the clinical practice guide- considering AD diagnoses evaluated and not evaluated lines among physicians in memory clinics in the ReDeGi by GPs, as sensitivity analysis: we included AD diagnoses catchment area.24 confirmed by the GPs or with evidence of treatment with We included patients recorded in ReDeGi between 2007 anti-dementia drugs in the numerator, and all AD diag- and 2016, and ascribed to primary care settings of the Catalan noses (evaluated or not by GPs) in the denominator. Using Health Service.The linkage was carried out using the national data from the linkage between SIDIAP and ReDeGi, we health identifier from the Catalan Health Service, which was For personal use only. estimated the PPV and sensitivity of AD cases. We calcu- available in both databases. We encoded the national health lated the PPV considering the number of AD cases identi- identifier and we used this ciphered identifier in all linkage fied in both ReDeGi and SIDIAP in the numerator, and procedures (Figure 1). Researchers did not have access to the the AD cases identified in SIDIAP, in the denominator. national health identifier or any other kind of identifying data. Sensitivity was defined as the number of AD cases identi- fied in both ReDeGi and SIDIAP in the numerator, and the AD cases identified in ReDeGi in the denominator. For Register of dementias of girona External data custodian both PPV and sensitivity, results were stratified by AD N=7,357 definition (Table 1), age, sex and severity. We provided Chiper ID and file code Password: email 95% CI assuming a normal distribution for all estimates. 2,391 excluded cases The level of significance was defined as p-value=0.05. All HTTPS (outside the CHS analyses were performed using R software v3.5.2.25 catchment area). Results Catalan Health Service (CHS) Survey sent to GPs Trusted 3rd party We sent the survey to the 42 GPs who agreed to parti- Decipher ID and linkage cipate, of which 29 (69%) provided feedback. The quota SIDIAP of patients assigned to these 29 GPs is described in Primary care data CHS custodian N=5.8 million Table S1. Of 188 patients with a diagnosis of AD, 137 Password: phone Anonymization and file code were evaluated within the 1-month response period (Figure 2). Table 2 provides a description of patient OwnCloud server characteristics. Of the diagnoses evaluated by GPs using the survey, 123 were considered as AD cases, and 14 were not con- firmed as AD cases (Figure 2). We estimated the PPV in Validation study research team Anonymized linked data (N=4,966) 89.8% (95% CI: 84.7–94.9). The supplementary materials show specific numerators and denominators used to esti- Figure 1 Register of dementias of Girona (ReDeGi) and SIDIAP linkage flow mate PPVs. Sensitivity analysis considered the 137 AD diagram. Data confidentiality measures were applied during information transfer diagnoses evaluated by GPs using the survey questions (encoded data in grey arrows; password in dotted grey lines). Abbreviation: SIDIAP, Information System for Research in Primary Care. and the 51 not evaluated, of which 32 were treated submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 511 DovePress Powered by TCPDF (www.tcpdf.org) Ponjoan et al Dovepress Online survey sent to 42 GPs 29 GPs answered. Dementia cases attended by participating GPs N=374 Excluded (n=119): other dementia than AD Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 AD cases attended by participating GPs N=188 Excluded (n=51): no reply within the 1-month response period AD cases evaluated by GPs N=137 Diagnosis by Diagnosis by test Evolved or incorrect specialist criteria diagnosis (N=127) (N=7) (N=3) For personal use only. AD cases, N=117 AD cases, n=6 AD cases, n=0 Not AD cases, N=10 Not AD cases, N=1 Not AD cases, N=3 Figure 2 Flow of Alzheimer's disease (AD) diagnoses evaluated by general practitioners (GPs) responding to an online survey. Table 2 Baseline demographic and clinical characteristics of patients according to the validation method: survey of general practitioners (GPs) to evaluate recorded AD diagnoses or linkage between datasets from primary (SIDIAP) and secondary care (ReDeGi) Survey Dataset Evaluated patients (n=137) Not-evaluated patients (n=51) All selected patients (n=4,966) Age, X (sd) 80.9 (7.7) 83.5 (6.9) 80.2 (6.9) Women, n (%) 92 (67%) 36 (70%) 3151 (63%) Deprivation index, n (%) Quintile 1: less deprived 12 (9%) 1 (2%) 629 (22.64) Quintile 2 32 (23%) 14 (27%) 472 (16.99) Quintile 3 21 (15%) 8 (16%) 545 (19.62) Quintile 4 34 (25%) 13 (26%) 653 (23.51) Quintile 5: more deprived 38 (28%) 15 (29%) 479 (17.24) Antidementia drugs, n (%) 92 (67%) 32 (63%) 3275 (66%) Severity of dementia, n (%) Mild - - 2955 (60%) Moderate - - 1532 (31%) Severe - - 437 (9%) submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 512 DovePress Powered by TCPDF (www.tcpdf.org) Dovepress Ponjoan et al patients. Sensitivity analysis provided a PPV of 82.4 (95% When the survey to GPs was the gold standard, our CI: 77.0–87.8) (Supplementary materials). Stratified ana- PPV estimate was similar to or slightly higher than pre- lyses are shown in Table 3. viously reported in primary care studies on overall dementia4 or AD.10,28 In other words, patients with Linkage between SIDIAP and the registry an AD diagnosis in EHR were very likely to have the disease. When the linkage with ReDeGi was the gold of dementias standard, PPV estimates varied between 66% and 74% This linkage provided data for 4,996 patients, as described depending on the applied algorithm. The lowest PPV was in Table 2. A specific AD diagnosis was found for 2,728 Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 achieved when using algorithm A2, revealing that some patients according to ReDeGi, 2,603 in SIDIAP and 1,947 treated patients might not actually have AD. The indica- in both databases. tions for anti-dementia drugs (anticholinesterases and PPV and sensitivity of AD diagnoses were 74.8 (95% memantine) are relatively specific for AD, but also can CI: 73.1–76.4) and 71.4 (95% CI: 69.6–73.0), respectively be used for dementia with Lewy bodies or other (Supplementary materials). When using algorithm A2 conditions.3 Indeed, when we restricted the treated patients (where cases included diagnosed or treated patients), considered as AD (algorithm A3), our PPV improved, PPV declined and sensitivity increased; when using algo- likely because the exclusion of patients who had a code rithm A3 (excluding cases with previous conditions), PPV indicating other subtypes of dementia, previous cerebro- was unchanged and sensitivity increased compared to vascular disease or Parkinsonism decreased the false posi- algorithm A1 (Table 4). No differences were observed For personal use only. tives. For example, to consider previous cerebrovascular when estimates were stratified by age, but men had lower diseases might help discern between AD and vascular PPV estimates than women, and sensitivity decreased with dementia, in line with previous EHR-based epidemiologi- severity in all AD case definitions (Table 4). cal studies of AD prevalence.10 Our PPV estimates differed depending on the gold stan- Discussion dard, and were higher when using data from the survey (82– We provided a comprehensive overview of the accuracy 89%) than from the ReDeGi linkage (66–74%). These of AD diagnoses routinely recorded in primary care. Our differences were unlikely to be related with the different PPV and sensitivity estimates indicated high accuracy study populations (ie, Catalonia in the survey, and Girona of AD codes in both global and stratified analyses, similar province in the linkage), since clinical practices are homo- to the one observed in other conditions such as cancer16 or geneous within the primary care services of the Catalan cardiovascular disease.26,27 We enhanced the robustness of Health Service.16 The variability might be related to differ- our findings by using two gold standards: a survey to GPs ences in validation methods and in diagnostic procedures in and a linkage with a provincial register of dementia. primary and secondary care. We cannot discard that selec- Estimates differed depending on the gold standard, and tion bias affected PPV values based on the survey, since values were higher using data from the survey than from GPs enrolled voluntarily and not all AD diagnoses were the ReDeGi linkage. evaluated. However, sensitivity analysis provided similar Table 3 Positive predictive value (PPV) of Alzheimer's disease (AD) diagnoses recorded in SIDIAP using information provided by the survey of general practitioners as reference. Sensitivity analysis included all AD patients of participating general practitioners (whether their diagnoses were evaluated or not) Population n Main analysis n Sensitivity analysis TP FP PPV (95% CI) TP FP PPV (95% CI) Women 92 84 8 91.3 (85.5–97.1) 128 106 22 82.8 (77.0–87.8) Men 45 39 6 86.7 (76.7–96.6) 60 49 11 81.6 (71.9–91.4) <80 years 84 74 10 88.1 (79.9–94.2) 105 89 16 84.8 (77.9–91.6) ≥80 years 54 49 4 92.4 (87.8–100.5) 83 66 17 79.5 (70.8–88.2) Total 137 123 14 89.8 (84.7–94.9) 188 155 33 82.4 (77.0–87.8) Abbreviations: TP, true positive; FP, false positive; SIDIAP, Information System for Research in Primary Care. submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 513 DovePress Powered by TCPDF (www.tcpdf.org) Ponjoan et al Dovepress Table 4 Positive predictive value (PPV) and sensitivity of Alzheimer's disease (AD) diagnoses in SIDIAP using three algorithms to define AD cases and linkage with the register of dementias of Girona as reference Algorithm Group Subgroup TP FP FN PPV (95% CI) Sensitivity (95% CI) A1. Diagnosed patients Sex Women 1,365 398 528 77.4 (75.4–79.3) 72.1 (70.0–74.1) Men 582 258 253 69.3 (66.1–72.3) 69.3 (66.1–72.3) Age <80 years 843 285 275 74.9 (72.3–77.4) 75.4 (72.8–77.8) ≥80 years 1,104 371 506 74.8 (72.5–77.0) 68.6 (66.3–70.8) Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 Severity Mild 1346 386 434 77.7 (75.7–79.6) 75.6 (73.6–77.6) Moderate 472 220 263 68.2 (64.6–71.6) 64.2 (60.7–67.7) Severe 123 43 83 74.1 (66.9–80.2) 59.7 (52.9–66.2) Total population 1,947 656 781 74.8 (73.1–76.4) 71.4 (69.6–73.0) A2. Diagnosed or treated patients Sex Women 1,623 715 270 69.4 (67.5–71.3) 85.7 (84.1–87.2) Men 711 484 124 59.5 (56.7–62.2) 85.1 (82.6–87.4) Age <80 years 992 531 126 65.1 (62.7–67.5) 88.7 (86.7–90.4) ≥80 years 1,342 668 268 66.8 (64.7–68.8) 83.4 (81.4–85.1) Severity Mild 1585 703 195 69.3 (67.4–71.1) 89.0 (87.5–90.4) For personal use only. Moderate 590 391 145 60.1 (57.0–63.2) 80.3 (77.2–83.0) Severe 152 87 54 63.6 (57.3–69.4) 73.8 (67.4–79.3) Total population 2,334 1,199 394 66.1 (64.5–67.6) 85.6 (84.2–86.8) A3. Diagnosed or treated patients without previous conditions Sex Women 1,599 558 294 74.6 (72.7–76.4) 83.9 (82.2–85.5) Men 687 346 148 67.5 (64.5–70.3) 81.8 (79.0–84.3) Age <80 years 973 378 145 72.7 (70.3–75.0) 86.7 (84.6–88.6) ≥80 years 1,313 527 297 72.1 (69.9–74.1) 80.9 (78.9–82.7) Severity Mild 1552 519 228 74.9 (73.0–76.7) 87.2 (85.6–88.7) Moderate 569 280 166 67.0 (63.8–70.1) 77.4 (74.3–80.3) Severe 144 59 62 70.9 (64.3–76.7) 69.9 (63.3–75.8) Total population 2,286 905 442 72.3 (70.7–73.9) 83.3 (81.8–84.6) Abbreviations: TP, true positive; FP, false positive; FN, false negative; SIDIAP, Information System for Research in Primary Care. results, suggesting a low level of overestimation. In parallel, under-recording of AD might occur when GPs use free- some underestimation of the PPV values based on the text instead of ICD-10 codes to record AD diagnoses, do linkage may have occurred, because we used a - not discern between dementia subtypes because they con- conservative AD definition. Previous studies considered sider the diagnosis and treatment of dementia useless,29,30 diagnoses of unspecified dementia as AD cases,7 but we or identify general dementia in a first visit and specify the restricted this criterion to treated patients, following Imfeld subtype in a later visit.7 Our sensitivity estimates improved et al (2013). We also used pure AD cases recorded in when patients treated with anti-dementia drugs were ReDeGi as a gold standard, while other studies included included as AD cases (algorithms A2 and A3). Anti- mixed dementia in the AD case definition.7 In primary care dementia drugs might have been prescribed in primary databases, the mixed dementia code is seldom used;10 there- care without recording a diagnostic code for AD in the fore, if AD codes were used instead, the number of false EHR, or a diagnosis of AD might have been made in positives would be increased. secondary care or in a private clinic, with no record of Sensitivity estimates obtained using the linkage analy- diagnosis in the primary care database of the Catalan sis showed that more than two-thirds of AD cases had Health Service.31 received a diagnosis. Identification of AD cases – treated Our findings disagree with previous evidence suggesting or not – depends greatly on the GP’s role. Some level of that less than half of the expected number of patients with submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 514 DovePress Powered by TCPDF (www.tcpdf.org) Dovepress Ponjoan et al dementia are recognized in primary care.32 Centers with an improve diagnostic accuracy in men and in severe AD active approach to dementia diagnosis are likely to record cases. Algorithms combining information on diagnosis a higher proportion of community cases on their health care and treatment for AD might be a powerful tool in primary databases.33 Thus, the quality of records regarding subtypes care research. Sensitivity estimates differed depending on of dementia might substantially depend on the dementia the method used – survey or linkage with a register,-but policies implemented in each primary care setting, and were sufficiently high to suggest that the primary care data their implementation or modifications over time.3 For were accurate for research purposes. example, dementia diagnosis rates improved substantially Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 during 10 years in the UK.31 In the Catalan Health Service, Acknowledgments dementia is actively considered; eg, a cognitive test is We acknowledge all the general practitioners from the administered in about 25% of people aged at least 65 AGICAP network who participated in this study. We thank years – with or without memory complaints.6 Moreover, Eduardo Hermosilla for data management support, Dr Vilalta- most of the AD diagnoses recorded in SIDIAP were done Franch for his expertise in the diagnosis of dementia, and by a specialist, suggesting that primary and secondary care Elaine Lilly, Ph.D., for the revision of the English text. This are well coordinated in the Catalan Health System.34,35 work was supported by the Real world Outcomes across Finally, we succeeded in identifying AD cases using the AD spectrum for better care: Multi-modal data Access algorithms. The integration of data related to AD diag- Platform (ROADMAP) from the Innovative Medicines noses, treatment and comorbidities enhanced case identifi- Initiative (Grant Agreement number 116020). This project For personal use only. cation. We recommend that future EHR studies was also supported by clinical research grants from Carlos develop AD case algorithms depending on their research III Health Institute, within the Net for Research in Preventive objectives. Large prospective studies might benefit from Activities and Health Enhancement (RedIAPP RD16/0007/ the definition of AD cases as diagnosed patients to avoid 0004) framework, and from the Agency for Management of a biased risk estimate due to a large number of false University and Research Grants (2017 SGR 1146). positives.3 In epidemiological studies, such as prevalence estimates, a definition of AD cases as diagnosed or treated patients would ensure identification of the highest number Disclosure of individuals having the condition. The authors report no conflicts of interest in this work. We acknowledge several limitations. First, the PPV could be overestimated because the validation study was References carried out with GPs from the AGICAP network, who are 1. Brookmeyer R, Johnson E, Ziegler-Graham K, Arrighi HM. regularly involved in clinical trials and therefore they could Forecasting the global burden of Alzheimer’s disease. Alzheimers Dement. 2007;3(3):186–191. doi:10.1016/j.jalz.2007.04.381 tend to be more accurate in registering diagnosis in the 2. Martin Prince A, Wimo A, Guerchet M, et al. World Alzheimer report electronic history of patients. Second, GPs voluntarily 2015 the global impact of dementia an analysis of prevalence, inci- answered the survey, which could affect the representative- dence, cost and trends; 2015. Availble from: https://www.alz.co.uk/ research/WorldAlzheimerReport2015.pdf. Accessed October 14, 2018. ness of the study sample. Third, we used ReDeGi as gold 3. Wilkinson T, Ly A, Schnier C, et al. Identifying dementia cases with standard, but this is a register based on cases recorded in routinely collected health data: a systematic review. Alzheimers Dement. 2018. doi:10.1016/j.jalz.2018.02.016 secondary care settings and therefore, it might not capture 4. McGuinness LA, Warren-Gash C, Moorhouse LR, Thomas SL. all AD cases occurred in the Girona province. However, the The validity of dementia diagnoses in routinely collected electro- ReDeGi has proved to cover about 75% of the expected nic health records in the United Kingdom: a systematic review. Pharmacoepidemiol Drug Saf. 2017;2019:1–12. doi:10.1002/ cases of dementia in the area under surveillance.36 Even pds.4669 more, ReDeGi provides good-quality diagnoses made by 5. Zunzunegui Pastor MV, Del Ser T, Rodríguez LA, García YMJ, Domingo J, Otero PA. Demencia no detectada y utilización de los specialists in neurology, geriatrics, psychiatry or internal servicios sanitarios: implicaciones para la atención primaria medicine from outpatient consultation offices in specialized TT - non-detected dementia and use of the health services: impli- care or in the hospital memory clinics. cations for primary care. Aten Primaria. 2003;31(9):581–586. Availble from: http://pesquisa.bvsalud.org/portal/resource/pt/ibc- 29693. Conclusion 6. Ponjoan A, Garre-Olmo J, Blanch J, et al. Epidemiology of dementia: prevalence and incidence estimates using validated electronic health Our findings suggested that EHR from primary care were records from primary care. Clin Epidemiol. 2018. In prepara:217–228. accurate to identify AD cases, but efforts can be made to doi:10.2147/CLEP.S186590 submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 515 DovePress Powered by TCPDF (www.tcpdf.org) Ponjoan et al Dovepress 7. Van De Vorst IE, Vaartjes I, Sinnecker LF, Beks LJM, Bots ML, 23. Morris JC. The Clinical Dementia Rating (CDR): current version Koek HL. The validity of national hospital discharge register data on and scoring rules. Neurology. 2012;43(11):2412. doi:10.1212/ dementia: a comparative analysis using clinical data from a university wnl.43.11.2412-a medical centre. Neth J Med. 2015;73(2):69. Available from: http://www. 24. Turró-Garriga O, Calvó-Perxas L, Vilalta-Franch J, et al. Adherence njmonline.nl/getpdf.php?id=1540. Accessed December 18, 2018. to clinical practice guidelines during dementia work-up in a 8. Sommerlad A, Perera G, Singh-Manoux A, Lewis G, Stewart R, real-world setting: a study from the registry of dementias of Girona. Livingston G. Accuracy of general hospital dementia diagnoses in Callahan B, ed. J Alzheimer’s Dis. 2017;59(3):997–1007. England: sensitivity, specificity, and predictors of diagnostic accuracy doi:10.3233/JAD-170284 2008–2016. Alzheimer’s Dement. 2018;14(7):933–943. doi:10.1016/j. 25. R Development Core Team. R: A Language and Environment for jalz.2018.02.012 Statistical Computing. Found stat comput Vienna, Austria; 2018. 9. Solomon A, Ngandu T, Soininen H, Hallikainen MM, Kivipelto M, Availble from: https://www.r-project.org/. Accessed September 8, Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 Laatikainen T. Validity of dementia and Alzheimer disease diagnoses 2018. in Finnish national registers. Alzheimers Dement. 2014;10 26. Sundbøll J, Adelborg K, Munch T, et al. Positive predictive value of (3):303–309. doi:10.1016/j.jalz.2013.03.004 cardiovascular diagnoses in the Danish National patient registry: 10. Imfeld P, Brauchli Pernus YB, Jick SS, Meier CR. Epidemiology, co- a validation study. BMJ Open. 2016;6(11):e012832. doi:10.1136/ morbidities, and medication use of patients with Alzheimer’s disease bmjopen-2016-012832 or vascular dementia in the UK. J Alzheimers Dis. 2013;35 27. Kivimäki M, Batty GD, Singh-Manoux A, Britton A, Brunner EJ, (3):565–573. doi:10.3233/JAD-121819 Shipley MJ. Validity of cardiovascular disease event ascertainment 11. Cohen JF, Korevaar DA, Altman DG, et al. STARD 2015 guidelines for using linkage to UK hospital records. Epidemiology. 2017;28 reporting diagnostic accuracy studies: explanation and elaboration. BMJ (5):735–739. doi:10.1097/EDE.0000000000000688 Open. 2016;6(11). doi:10.1136/bmjopen-2016-012799 28. Jaakkimainen RL, Bronskill SE, Tierney MC, et al. Identification of 12. Benchimol EI, Smeeth L, Guttmann A, et al. The reporting of studies physician-diagnosed alzheimer’s disease and related dementias in conducted using observational routinely-collected health data population-based administrative data: a validation study using family (RECORD) statement. PLoS Med. 2015;12(10):e1001885. physicians’ electronic medical records. J Alzheimer’s Dis. 2016;54 doi:10.1371/journal.pmed.1001885 (1):337–349. doi:10.3233/JAD-160105 For personal use only. 13. Domínguez-Berjón MF. Construcción de un índice de privación 29. Lang L, Clifford A, Wei L, et al. Prevalence and determinants of a partir de datos censales en grandes ciudades españolas (Proyecto undetected dementia in the community: a systematic literature review MEDEA). Gac Sanit. 2008;22(3):179–187. and a meta-analysis. BMJ Open. 2017;7(e011146):1–8. doi:10.1136/ 14. Bolíbar B, Fina Avilés F, Morros R, et al. [SIDIAP database: electro- bmjopen-2016-011146 nic clinical records in primary care as a source of information for 30. Drummond N, Birtwhistle R, Williamson T, Khan S, Garies S, epidemiologic research]. Med Clin (Barc). 2012;138(14):617–621. Molnar F. Prevalence and management of dementia in primary care doi:10.1016/j.medcli.2012.01.020 practices with electronic medical records: a report from the Canadian 15. García-Gil MDM, Hermosilla E, Prieto-Alhambra D, et al. Construction primary care sentinel surveillance network. C Open. 2016;4(2):E177– and validation of a scoring system for the selection of high-quality data E184. doi:10.9778/cmajo.20150050 in a Spanish population primary care database (SIDIAP). Qual Prim 31. Donegan K, Fox N, Black N, Livingston G, Banerjee S, Burns A. Care. 2012;20(2):135–145. Available from: http://www.ncbi.nlm.nih. Trends in diagnosis and treatment for people with dementia in the UK gov/pubmed/22688222. Accessed June 22, 2012. from 2005 to 2015: a longitudinal retrospective cohort study. Lancet 16. Garcia-Gil M, Elorza J-M, Banque M, et al. Linking of primary care Public Heal. 2017;2(3):e149–e156. doi:10.1016/S2468-2667(17) records to census data to study the association between socioeco- 30031-2 nomic status and cancer incidence in Southern Europe: a Nation-wide 32. Connolly A, Gaehl E, Martin H, Morris J, Purandare N. Underdiagnosis ecological study. PLoS One. 2014;9(10):e109706. doi:10.1371/jour- of dementia in primary care: variations in the observed prevalence and nal.pone.0109706 comparisons to the expected prevalence. Aging Ment Health. 2011;15 17. Ramos R, Balló E, Marrugat J, et al. Validity for use in research on (8):978–984. doi:10.1080/13607863.2011.596805 vascular diseases of the SIDIAP (Information system for the devel- 33. Perera G, Pedersen L, Ansel D, et al. Dementia prevalence and opment of research in primary care): the EMMA study. Rev Española incidence in a federation of European electronic health record data- Cardiol. 2012;65(1):29–37. doi:10.1016/j.recesp.2011.07.017 bases: the European medical informatics framework resource. 18. Fina-Aviles F, Medina-Peralta M, Mendez-Boo L, et al. The descrip- Alzheimer’s Dement. 2018;14(2):130–139. doi:10.1016/j.jalz.20 tive epidemiology of rheumatoid arthritis in Catalonia: a retrospective 17.06.2270 study using routinely collected data. Clin Rheumatol. 2016;35 34. Vilalta-Franch J, López-Pousa S, Calvó-Perxas L, Garre-Olmo J. (3):751–757. doi:10.1007/s10067-014-2801-1 Psychosis of Alzheimer disease: prevalence, incidence, persistence, 19. Pujades-Rodriguez M, Assi V, Gonzalez-Izquierdo A, et al. The diag- risk factors, and mortality. Am J Geriatr Psychiatry. 2012. nosis, burden and prognosis of dementia: a record-linkage cohort study doi:10.1097/JGP.0b013e31826ce617 in England. PLoS One. 2018. doi:10.1371/journal.pone.0199026 35. Vilalta-Franch J, Garre-Olmo J, López-Pousa S, Llinàs-Reglà J, 20. Vinyoles Bargalló E, Pujol Ribó C, Red AGICAP. [Clinical trials in Calvó-Perxas L, Cubí-Montfort R. [Feasibility of a telemedicine primary care. The experience of the AGICAP network in Catalonia]. support system for diagnosing dementia in primary care]. Atención Primaria. 2004;34(1):38–42. Rev Neurol. 2012;55(5):263–269. Available from: http://www. 21. Garre-Olmo J, Flaqué M, Gich J, et al. A clinical registry of dementia ncbi.nlm.nih.gov/pubmed/22930137. Accessed December 16, based on the principle of epidemiological surveillance. BMC Neurol. 2014. 2009;9:5. doi:10.1186/1471-2377-9-5 36. Calvó-Perxas L, Aguirregomozcorta M, Casas I, et al. Rate of 22. German RR, Lee LM, Horan JM, et al. Updated guidelines for evaluating dementia diagnoses according to the degree of aging of public health surveillance systems: recommendations from the Guidelines the population. Int Psychogeriatrics. 2015;27(03):419–427. Working Group. MMWR Recomm Rep Morb Mortal Wkly Rep Recomm doi:10.1017/S1041610214002130 Rep. 2001;50(RR–13):1–35; quiz CE1–7. Available from: http://www. ncbi.nlm.nih.gov/pubmed/18634202. Accessed June 22, 2012. submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 516 DovePress Powered by TCPDF (www.tcpdf.org) Dovepress Ponjoan et al Supplementary materials ● When using data from the linkage between the Register of Dementias of Girona (ReDeGi) and Information Estimates of positive predictive value (PPV) and System for Research in Primary Care (SIDIAP) (algo- sensitivity of Alzheimer’s disease (AD) diagnoses rithm A1): recorded in primary care ● When using data from the online survey of general 1947 AD cases identified in both practitioners (GPs): ReDeGi and SIDIAP PPV ¼  100 ¼ 74:8 2603 AD cases identified in 123 confirmed diagnoses of AD PPV ¼  100 SIDIAP Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 137 AD diagnoses evaluated by GPs ¼ 89:8 Sensitivity analysis: considering AD diagnoses evaluated 1947 AD cases identified in both ReDeGiand SIDIAP and not evaluated by GPs. Sensitivity ¼  100 ¼ 71:4 2728 AD cases identified 123 confirmed diagnoses of AD in ReDeGi þ32 patients treated with antidementia drugs PPV ¼  100 137 evaluated AD diagnoses þ51 not evaluated AD diagnoses ¼ 82:4 For personal use only. Question: At this point, what is the basis for a diagnosis of dementia? Answer 1- The diagnosis was made by a hospital specialist Indicate the subtype of dementia Indicate the severity of dementia Answer 2- The diagnosis was based on cognitive and functional tests Specify the test(s) and the score(s) Indicate the subtype of dementia Indicate the severity of dementia Are the DSM-IV or ICD-10 diagnostic criteria fulfilled? (Yes/No) Answer 3- The diagnosis was based on a clinical opinion symptoms were observed by a general practitioner, care giver or patient relative. Indicate the subtype of dementia Indicate the severity of dementia Are the DSM-IV or ICD-10 diagnosis criteria fulfilled? (Yes/No/Not sure) Answer 4- The diagnosis is null (evolved or incorrect diagnosis) Answer 5- Uncertain due to lack of information Figure S1 Summary of survey for the validation of dementia diagnosis. Abbreviations: DSM-IV, Diagnostic and Statistical Manual of Mental Disorders, 4th edition; ICD-10, International Statistical Classification of Diseases and Related Health Problems, 10th edition. submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 517 DovePress Powered by TCPDF (www.tcpdf.org) Ponjoan et al Dovepress Table S1 Description of the quota of patients from the 29 general practitioners who answered the survey Sociodemographic characteristics Quota of patients (n=29) Percentage of men, X (sd) 49.5 (3.4) Age, X (sd) 47.5 (2.0) Percentage of people aged ≥65 years, X (sd) 21.4 (4.9) Number of patients, X (sd) 1426 (331) Rural areas, n (%) 5 (17.1) Socioeconomic status in urban areas, n (%) Clinical Epidemiology downloaded from https://www.dovepress.com/ by 80.174.193.252 on 04-Aug-2021 Status 1: less deprived 2 (8.3) Status 2 2 (8.3) Status 3 6 (25) Status 4: more deprived 14 (58.3) For personal use only. Clinical Epidemiology Dovepress Publish your work in this journal Clinical Epidemiology is an international, peer-reviewed, open access, systematic reviews, risk & safety of medical interventions, epidemiol- online journal focusing on disease and drug epidemiology, identifica- ogy & biostatistical methods, and evaluation of guidelines, translational tion of risk factors and screening procedures to develop optimal pre- medicine, health policies & economic evaluations. The manuscript ventative initiatives and programs. Specific topics include: diagnosis, management system is completely online and includes a very quick prognosis, treatment, screening, prevention, risk factor modification, and fair peer-review system, which is all easy to use. Submit your manuscript here: https://www.dovepress.com/clinical-epidemiology-journal submit your manuscript | www.dovepress.com Clinical Epidemiology 2019:11 518 DovePress Powered by TCPDF (www.tcpdf.org)