than that used by Perera et al, who identified cases using only diagnostic codes or free text.7 Demographic aspects might also contribute to explaining differences between primary-care databases from different countries. Life expectancy in Spain is the longest in the EU (83.3 years),47 and the sex gap in life expectancy favors women to a higher extent than in other countries, such as Denmark, the Netherlands, or the UK.47 Since dementia prevalence is higher in women and increases with age, demographic characteristics of the population might facilitate a higher number of dementia cases in Spain than in other European countries. Primary-care characteristics might also contribute to explaining disparities between studies on dementia epidemiology based on electronic health records from primary care. Catalan primary-care settings from which SIDIAP obtained data might be highly sensitive in diagnosing dementia cases. On one hand, cognitive tests are administered widely to identify dementia cases in the Catalan primary-care system: in SIDIAP, >22% of patients with no dementia code in their electronic health history had taken a cognitive test. Patients with memory complaints reported by themselves, their relatives, or general practitioners are asked to take a cognitive test. Patients included in specific programs for home care or institutionalized due to chronic complex disease may also be asked to complete a cognitive test, even if they have no memory or cognitive complaints. The huge number of cognitive tests administered in Catalan primary care might facilitate the identification of new dementia cases.14 On the other hand, we found that most patients with a dementia code in SIDIAP had been diagnosed by a specialist or prescribed antidementia drugs by specialists. These findings suggest a good coordination between primary and secondary settings provided by the Catalan Health Institute, strengthened by such initiatives as the Computerized Support System for the Diagnosis of Dementia in Primary Care (SISDDAP [Catalan acronym]), an asynchronous telemedicine program involving both primary- and specialized-care professionals that establishes protocols for the screening and diagnosis of dementia.48 Appropriate communication and coordination between primary and secondary care might facilitate the identification of new dementia cases and the recording of diagnoses in the electronic health record. Our findings are in line with previous literature reporting that settings with active dementia-diagnosis centers are likely to see a higher proportion of community cases in health care databases.7 Our findings also suggested that the high level of underdiagnosis of dementia reported in several primary-care systems31,49 might not be true for other primary-care settings. General practitioner concern and training, as well as programs and policies designed to better diagnose dementia cases, might help to reduce underdiagnosis in primary care. Therefore, databases containing routinely collected data from primary care might be a powerful tool to study dementia epidemiology if data have been recorded in a setting with active management of dementia cases and validated as sufficiently accurate for research purposes. Limitations and strengths We acknowledge several limitations. First, the PPV could have been overestimated, because the validation study was constrained to general practitioners from the AGICAP network, who are regularly involved in clinical trials and thus could be more prone to register diagnoses accurately in electronic medical records. Second, as not all dementia diagnoses were evaluated by general practitioners, we cannot discard a potential bias that might have resulted in an overestimation of the PPV. Therefore, we replicated the PPV estimate considering the worst-case scenario, in which data from both evaluated and unevaluated dementia diagnoses were used. This worstcase scenario PPV estimate would indicate the minimum value of the PPV, ie, if bias had occurred and the PPV value had been overestimated, the PPV would not be lower than the worst-case scenario estimate. Third, the validation study did not provide sensitivity estimates