Incidence and Prevalence of Dementia: A 2015–2020 Population-Based Study in the Campania Region of Italy Open Access

Dementia refers to a decline in cognitive abilities, which becomes substantial enough to disrupt individuals’ ability to function independently in their daily activities [1]. Dementia develops as a result of complex interplays between genetic, lifestyle, and environmental factors [2]. The most common cause of dementia is deemed to be Alzheimer’s pathology, followed by vascular pathology and by a number of other neurodegenerative diseases.

Dementia is the seventh leading cause of death globally and a major cause of dependency and disability. As populations age, the number of people affected by dementia is expected to grow, and in Europe, the number of dementia cases is estimated to increase from 7.7 million in 2001 to 15.9 million in 2040 [3, 4]. Current estimates account for more than 55 million people with dementia worldwide [5]. Living with dementia can be harrowing for individuals and their families, and the social and economic consequences are challenging for all societies and care systems [6]. Direct medical costs, direct social sector costs (including long-term care), and costs of informal care are enormous, distributed inequitably, and substantially greater than previous estimates [7, 8].

Considering the high societal and financial burden of dementia, along with the possibility to implement preventative strategies to identify and address modifiable lifestyle and environmental factors [2], population-based estimates of dementia incidence and prevalence are crucial for public health planning and resource prioritisation. To date, only few studies have exclusively employed administrative data to assess the epidemiological burden of dementia. Furthermore, most of these studies lack of formal validation, particularly in the context of sensitivity and specificity analyses or rely on data characterised by a reduced level of granularity [9, 10]. Thus, in the present study, we aimed to (1) estimate the 2015–2020 prevalence and incidence of any dementia and Alzheimer’s dementia (AD) using routinely collected healthcare data in the Campania region (South Italy); (2) estimate the proportion of undetected cases; and (3) validate our algorithms for case identification towards a clinical registry.

This is a population-based study, obtained from the retrospective analysis of routinely collected healthcare data from the whole population living in the Campania region (South Italy) from 2015 to 2020 (population on January 1, 2021; 5,712,143 with 2,784,616 males and 2,927,527 females) (http://dati.istat.it/). The study was approved by the Federico II Ethics Committee (332/21). All patients signed informed consent authorising the use of anonymised, routinely collected healthcare data, in line with data protection regulation (GDPR EU2016/679). The study was performed in accordance with good clinical practice and the Declaration of Helsinki.

The dataset was created by merging different data sources of the Campania region. The cohort comprised all individuals residing in Campania region aged ≥65 years alive at the prevalence day (January 1, 2021) who had at least one record from the following databases.

• 1.

  Hospital Discharge Record database, which included all admissions in the study period with an ICD-9 CM code of dementia as one of the discharge diagnoses (online suppl. Material 1; for all online suppl. material, see https://doi.org/10.1159/000539031) [4, 11, 12]

• 2.

  Regional Drug Prescription database, which included all medications prescribed to people aged ≥65 years likely due to dementia (e.g., aripiprazole, clozapine, donepezil, galantamine, memantine, olanzapine, quetiapine, risperidone, rivastigmine, ziprasidone) [13, 14]

• 3.

  Exemptions database with dementia-specific exemptions (e.g., AD, another dementia)

Hospital admissions and drug prescriptions delivered out of the Campania region are reported to the Campania Region Healthcare Regulatory Society (So.Re.Sa.) for refund purposes and then included in the above-mentioned datasets. Our data sources encompass both active components (such as drug prescriptions and hospitalisations) and passive components (such as exemption codes for disability benefits) of healthcare utilisation over 6 years, enabling us to include individuals who may not regularly seek healthcare services.

From the database, individuals with a diagnosis of dementia not living in the Campania region were filtered out. Patient unique identifier code was fully anonymised by the Campania region Healthcare Regulatory Society (So.Re.Sa.) before releasing the datasets. Data were fully anonymised by the Campania Region Healthcare Regulatory Society (So.Re.Sa.) before releasing the datasets. As the same anonymisation algorithm was used across datasets, data merging was possible.

We classified individuals as having dementia in the case of at least one record in the previously mentioned datasets (any dementia diagnosis in hospital discharge records; prescriptions of any dementia-related treatments; and any dementia exemptions). Then, we further classified individuals as having AD in the case of records consistent with AD in the previously mentioned datasets (AD diagnosis in hospital discharge records; prescriptions of donepezil, galantamine, memantine, and/or rivastigmine; and AD exemption).

For a subset of the included population, we performed linkage to the clinical registry at the Cognitive Impairment and Neurorehabilitation Unit. This clinical registry was selected due to ease of ethical clearance (because it originates from the same institution of administrative healthcare data handling) and to the large number of individuals referred yearly.

In particular, we included all individuals tested during the study period (2015–2020) and extracted neuropsychological test results including the mini-mental status examination (MMSE), along with more specific tests for different cognitive domains (e.g., memory, attention and executive, visuospatial and abstract reasoning). Test results were age-, sex-, and education-adjusted using Italian normative values. Based on our algorithms for the identification of any dementia and AD, we classified individuals with available neuropsychological test results as having any dementia and/or AD in routinely collected healthcare data.

Missing data were present for birth date (<1% of the population). We used multiple imputation by chained equations (10 copies) to estimate missing data, aiming to preserve the relationships between variables and capture the uncertainty associated with imputed values. The missing data were estimated based on sex, town of residence, year of the first record in the administrative database, and Charlson Comorbidity Index.

To estimate prevalence (aim 1), age-and sex-standardised prevalence rates were calculated for the whole cohort and then stratified by sex using the direct standardisation method. The European population in 2020 was considered as the reference population [15] because the similarity of health systems and standards among European countries makes it easier to compare health indicators and disease rates between different populations. The prevalence rate was expressed as the number of cases per 1,000 inhabitants. To calculate 95% confidence intervals (95% CI) for the standardised rates, the Byar’s approximation method based on the Poisson distribution was used.

To estimate the incidence (aim 1), we calculated the minimum time interval between our study baseline (January 1, 2015) and the index date (hypothesising the index date was the first contact with the healthcare system), using the clinical dataset to validate the algorithm. We tested three possible algorithms, which differed for the duration of the time interval between study baseline and index date: (1) at least 12 months from January 1, 2015; (2) at least 24 months from January 1, 2015; and (3) at least 36 months, from January 1, 2015. We assessed the discrimination power of the candidate algorithms (using a time frame of 12, 24, and 36 months from January 1, 2015) to identify incident cases by calculating sensitivity, specificity, positive and negative predictive values, as well as area under the curve (AUC). Once we identified the most accurate algorithm, we calculated cumulative dementia incidence rate for the years 2015–2020, annual dementia incidence rate, and 95% CIs. The incidence rate was expressed as the number of new cases per 100,000 inhabitants, and 95% CIs were calculated using Byar’s approximation of the Poisson test.

To estimate the proportion of undetected cases (aim 2), we used the capture-recapture method (keeping the boundary conditions of sampling unchanged to improve the accuracy). Model selection was based on Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and goodness-of-fit-based CIs following the method published by Regal and Hook [16].

To validate our algorithms towards the clinical registry (aim 3), we employed logistic regression models to estimate associations between different neuropsychological test results (independent continuous variables) and binary disease status based on routinely collected healthcare data (dependent variable, e.g., any dementia vs. no dementia; AD vs. no dementia). Results are presented as coefficients, 95% CIs, and p values. Based on these models, we obtained AUCs, using different neuropsychological test results as the main explanatory variables. Then, we calculated sensitivity, specificity, percentage of correctly classified individuals, likelihood ratio for a positive test result, and likelihood ratio for a negative test result of our algorithms for any dementia and/or AD towards the identification of individuals with MMSE ≤24 or MMSE ≤26. To compare sex and age between individuals identified as having dementia from the whole population with those from the clinical registry, we employed the χ² test and t test, respectively.

Statistical analyses were performed using Stata 17.0. Results were considered statistically significant for p < 0.05.

Out of 1,118,545 individuals older than 65 years living in the Campania region of Italy (among 5,624,260 inhabitants as of January 1, 2021), we identified 80,392 people older than 65 years with dementia (females = 62.4%; age = 80.5 ± 7.6 years), among which 35,748 people older than 65 years specifically fit our criteria for diagnosing AD (females = 63.0%; age = 79.2 ± 6.4 years). The proportion of patients identified from different data sources (hospital discharge records, drug prescriptions (Table 1), and exemptions) is presented in Figure 1.

Age and sex distributions are showed in Figure 2. Age- and sex-standardised prevalence rate of any dementia per 1,000 people was 77.64 (95% CI = 77.57; 77.68) (85.22 for females [95% CI = 85.14, 85.24] and 67.54 for males [95% CI = 67.46, 64.62]). The age- and sex-standardised prevalence rate of AD per 1,000 people was 34.05 (95% CI = 34.01; 34.09) (37.63 for females [95% CI = 37.59, 37.69] and 29.28 for males [95% CI = 29.24, 29.34]).

We developed three different algorithms based on the interval between baseline (January 1, 2015) and the index date (first dementia-related record in routinely collected healthcare data). We favoured the algorithm identifying individuals with at least 36 months of lack of data for any dementia and the algorithm identifying individuals with at least 24 months of lack of data for AD. The algorithm for any dementia (at least 36 months as the characterisation period) included 28,873 incident dementia cases, with 0.79 sensitivity (95% CI = 0.67, 0.91), 0.62 specificity (95% CI = 0.54, 0.71), 0.71 ROC area (95% CI = 0.67, 0.74), 0.59 positive predictive value (95% CI = 0.51, 0.69), and 0.81 negative predictive value (95% CI = 0.70, 0.93). The algorithm for AD (at least 24 months as the characterisation period) included 17,551 incident AD cases, with 0.80 sensitivity (95% CI = 0.70, 0.92), 0.59 specificity (95% CI = 0.48, 0.72), 0.70 ROC area (95% CI = 0.65, 0.74), 0.76 positive predictive value (95% CI = 0.66, 0.87), and 0.65 negative predictive value (95% CI = 0.53, 0.79). Other tested algorithms did not show significant improvements in sensitivity, specificity, and AUC (online suppl. Material 2).

The cumulative incidence obtained for any dementia and AD was 492.58 (95% CI = 486.9, 498.3) and 299.43 (95% CI = 295.01, 303.89) per 100,000 from 2015 to 2020, respectively. The annual incidence of dementia and AD for the Campania region was 82.10 (95% CI = 81.15, 83.05) and 59.89 (95% CI = 59.00, 60.78) per 100,000 respectively (Fig. 3).

In the capture-recapture analysis, the model with the best fit showed that the number of expected cases was 80,499 for any dementia (95% CI = 80,392; 119,935) and 53,024 for AD (95% CI = 45,838; 66,492), accounting for a 1.7% and 3.0% increase, respectively, compared with the number of detected cases.

We obtained neuropsychological test results of 3497 individuals who were referred to our Cognitive Impairment and Neurorehabilitation Unit during the study period (2015–2020). When comparing people identified as having dementia using our algorithm with those for whom neuropsychological test results are available (Fig. 4), we found similar distribution in sex (χ² = 0.4, p = 0.51) but younger age for those with neuropsychological test results (t = 21.7; p < 0.01).

We observed significantly lower neuropsychological test scores for all cognitive domains in individuals identified as having any dementia and/or AD in routinely collected healthcare data, when compared with individuals without records of dementia in hospital discharge records, drug prescriptions, and exemptions (Table 2). Overall, our algorithms showed acceptable performance with AUC ranging from 0.59 to 0.72 and double likelihood ratio of correctly identifying individuals above and below MMSE standard cut-offs (24 and 26) (Table 3) [17].

We provided up-to-date (2015–2020) population-based estimates of prevalence and incidence of any dementia and AD, using routinely collected healthcare data from a representative European area (Campania region, South Italy), and validated towards a clinical registry. Population-based cohorts are deemed to be a cornerstone for policy evaluation in dementia [18], with most evidence so far derived from individual-level data. As such, this paper sets the background for using our validated cohort, including a combination of administrative and clinical data, for assessing risk factors, prevention policies, and inequalities in dementia. Thanks to the validation results, our algorithm can be as well applied in other settings, considering the similarities in data structure in other Italian regions and European Health Systems.

In our cohort of individuals older than 65 years, age- and sex-standardised prevalence rate was 7.7% for any dementia and 3.4% for AD. Our prevalence rates fall within the expected intervals for Western countries [4, 19, 20] and are in line with Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2019 estimates [4]. In a meta-analysis including 9 studies that were carried out in Europe between 1993 and 2018 using DSM-IV criteria for diagnosis [21], the age- and sex-standardised prevalence rate was 7.1%. In a more recent German study (2014–2019) [22], prevalence of dementia was slightly higher (10.3% in individuals over 65 years), suggesting that prevalence has been increasing over time, as from a higher incidence and/or reduced mortality. Among individuals with dementia, in our cohort, AD possibly accounted for about 50% of the cases, with 3.4% prevalence over the age of 65 years. Again, this is in line with a previous study in England, using routinely collected healthcare data (from the Clinical Practice Research Datalink), that estimated 3.7% prevalence of diagnosed AD [4]. However, clinicians may encounter challenges in accurately identifying different forms of dementia and thus might refer to AD for ease to access to medications and health benefits, with subsequent epidemiological overestimates [23]. We also found higher prevalence rates of any dementia and AD in females than males (about 1.3 female-to-male ratio); this has already been described by other epidemiological studies, with up to 1.69 female-to-male ratio [4]. Similarly, the age of our population is in line with a previous Norwegian study (about 80 years old) [12].

The annual incidence of any dementia and AD for the Campania region were 82.10 and 59.89 per 100,000, respectively. These estimates are in line with 2019 estimates of the global incidence of dementia (95.0 per 100,000 persons) [24]. As such, AD accounts for 50% of prevalent cases and 70% of incident cases, possibly suggesting that the progressive uptake of pathology-based diagnosis of dementia has made the diagnosis of AD more precise and frequent in patients newly presenting with symptoms suggestive of dementia [25], when compared with historical cohorts.

Our algorithm identified individuals living with dementia from the combination of drug prescriptions, exemption codes, and hospital discharge records, and thus, its design focused on detecting diagnosed cases rather than pre-clinical manifestations or cases without formal diagnosis. The methodology employed in this study was in line with the prior literature using drug prescriptions, hospital discharge records, and exemptions [10, 26], though with a notable strength related to the use of clinical datasets for algorithm validation. We developed our algorithms to prioritise higher sensitivity over specificity, addressing the pervasive global issue of underdiagnosis in dementia cases [27‒29]. We had 0.79 sensitivity for any dementia (using at least 36 months as the characterisation period) and 0.80 for AD (using at least 24 months as the characterisation period). As such, the detection of AD seems easier and more feasible, than the identification of other forms of dementia, possibly because of the availability of Alzheimer’s pathology biomarkers [25]. Notwithstanding, most previous studies have focused on dementia, few have included AD, and none have provided actual population-based estimates of other types of dementia. Also, looking at neuropsychological performance, our algorithms were able to identify individuals with more impaired cognitive function across all explored domains. Only three previous studies have used a clinical registry as the reference standard for calculating sensitivity and specificity. The estimated sensitivity in our study for any dementia (79%) and for AD (80%) is higher than the Finnish Cardiovascular Risk Factors, Aging and Dementia (CAIDE) study (64% for dementia and 71% for AD) and the Australian Concord Health and Ageing in Men Project (CHAMP) study (20% for dementia) [30, 31]. When compared with the US ADAMS study (Aging, Demographics, and Memory Study) using Medicare claims (inpatient hospital claims, home health, and the use of a skilled nursing facility), we performed similarly in identifying dementia (79% vs. 85%) and much better for AD (80% vs. 64%) [32]. Taken together, our approach, using high granularity data including both active and passive components of healthcare utilisation, performed much better than previous studies and, based on capture and recapture models, had very low rates of undetected cases (1.7% for any dementia and 3.0% for AD).

Most individuals with dementia were identified through drug prescriptions that were a combination of medications enhancing memory function and anti-psychotic medications [14], thus addressing the combination of cognitive and behavioural changes associated with dementia [22]. There is only one previous study describing medication utilisation in dementia using routinely collected data showing that donepezil was the most frequently prescribed medication in the US and the UK [13], while memantine was preferred in our population (followed by donepezil and rivastigmine). On the contrary, this is the first study describing the use of anti-psychotic medications in dementia clinical practice, which will warrant further investigation. Of course, the use of antipsychotics to identify people with dementia in administrative records can lead to an overestimation of the prevalence of dementia, due to diverse utilisation in clinical practice. However, most of the prescriptions in the dataset were for AD-specific drugs, and only 30% of the included population with dementia was identified uniquely based on the prescription of antipsychotics. Furthermore, independently from the main driver of the prescription, people aged 65 years or older who use antipsychotics are very unlikely to have normal cognitive function [33]. Additional sources of patient identification were hospital discharge records, where a diagnosis of dementia could be included only in the presence of consultations from dementia specialists and/or related healthcare resource utilisation, thus minimising the risk of false positive results. Finally, a subset of patients was identified using exemption codes that grant copayment benefits; however, the same benefits can be granted through different exemption codes (e.g., other comorbidities, low income), and thus, dementia exemption code is not always needed to patients.

Interestingly, the majority (69%) of the population referred to neuropsychological testing did not have a diagnosis of dementia in the administrative records. These individuals may actually have subjective cognitive complaints or mild cognitive impairment and might be a candidate for novel pathogenetic therapy in the upcoming years and/or to preventative policies. Of note, individuals with dementia who had neuropsychological tests were younger than those identified uniquely in routinely collected healthcare data, thus suggesting that neuropsychological tests are taken in the early stages of the disease [34]. Despite the presence of a large number of Cognitive Disorders and Dementia Centres (CDCDs) operating in the Campania region [35], the clinical registry covers approximately 10% of the regional population with dementia and holds similar distribution in sex.

We have to acknowledge that the study was conducted in a limited geographical area. Nevertheless, the Campania region, with approximately 6 million inhabitants, ranks as the second most populated region in Italy [36]. Also, we have focused on dementia epidemiology (incidence, prevalence), while we did not assess a number of additional factors that potentially affect our outcomes (e.g., comorbidity burden) [37, 38]. We specifically focused on individuals aged 65 years and older, to ensure comparability with previous studies, which predominantly focused on this age group. Considering that there is a trend of dementia onset occurring at younger ages, in the future, we will need to consider more stringent criteria to differentiate dementia at any age from psychiatric and/or neurodevelopmental disorders, as well as to identify the early/pre-clinical manifestations of dementia. Our cohort was set in 2015, and thus, follow-up does not allow calculation of temporal trends of dementia [11], which is warranted for the future. Diagnosis of dementia and in particular, of AD is based on the combination of clinical and pathology measures (e.g., CSF biomarkers, PET) [25], while we relied on healthcare records, independently of test results, as in population-based studies [37]. However, population-based approaches have the potential to result in substantial incidence and prevalence reductions, greater health equity, and longer term effects, than the individual level and more clinically granular approaches [18]. Also, population-based studies do not hold representative bias that has been affecting most dementia research [39].

In conclusion, we estimated the 2015–2020 prevalence and incidence of any dementia and AD using routinely collected healthcare data in the Campania region (South Italy) and found rates and population characteristics in line with previous similar studies. Our algorithms were validated towards a clinical registry and, thanks to the combination of administrative and neuropsychological data, could be used in the future to assess the epidemiological burden of dementia, thereby supporting the identification of risk factors, the planning of healthcare access, and the development of effective prevention strategies.

The study was approved by the Federico II Ethics Committee (332/21). All patients signed informed consent authorising the use of anonymised, routinely collected healthcare data, in line with data protection regulation (GDPR EU2016/679). The study was performed in accordance with good clinical practice and the Declaration of Helsinki.

Marcello Moccia has received research grants from the ECTRIMS-MAGNIMS, the UK MS Society, and Merck; honoraria from Biogen, BMS Celgene, Ipsen, Merck, Novartis, Roche, and Sanofi-Genzyme. Vincenzo Brescia Morra has received research grants from the Italian MS Society and Roche and honoraria from Bayer, Biogen, BMS Celgene, Merck, Mylan, Novartis, Roche, Sanofi-Genzyme, and Teva. Raffaele Palladino has received research grants from Sanofi-Genzyme. Other authors have nothing to disclose.

This work was financially supported by the MUR PNRR Extended Partnership (MNESYS No. PE00000006) to Marcello Moccia. Additionally, funding was provided by the Campania region through the E65E22000370002 funding source to Raffaele Palladino. The work was also financially supported in part by the Italian Ministry of Health, through the project HubLife Science – Digital Health (LSH-DH) PNC-E3-2022-23683267 – DHEAL-COM – CUP E63C22003790001, within the “National Plan for Complementary Investments – Innovative Health Ecosystem” – Unique Investment Code: PNC-E.3 to Marcello Moccia, Raffaele Palladino, and Maria Triassi. The funders played no role in data acquisition, analysis, interpretation, and publication.

R.P. and M.M. conceived and designed the study. G.A. analysed the data. G.A., M.T., V.B.M., and R.P. discussed the data analyses and interpreted the results. G.A., M.M., and R.P. wrote the first draft of the manuscript. All the authors critically revised the manuscript, approved the final manuscript for publication, and agreed to act as guarantors of the work. G.A. has full access to all data used in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All the authors have read and agreed to the published version of the manuscript.

The data presented in this study are available on request from the corresponding author. The data are not publicly available.