The unprecedented burden on healthcare systems caused by the coronavirus disease 2019 (COVID-19) pandemic has highlighted the need for tools to assist decision-makers in planning resource allocation. Since the first wave of the pandemic, healthcare providers have focused on the need to promptly identifying risk factors for COVID-associated severe disease and death (1). Age and comorbidity have been widely adopted as the main, if not the only, criteria to allocate healthcare resources and stratify care levels in affected individuals (2, 3). However, these parameters do not fully capture the biological heterogeneity of individual health trajectories, especially in older patients. Indeed, there is significant variability in socio-economic, functional, and cognitive status across individuals, as well as in lifestyle habits.

Frailty is a clinical syndrome that can be defined as a decline in body functions and physiological reserves, resulting in increased vulnerability and reduced resilience to physical and mental stressors, ultimately leading to an increased risk of negative health outcomes (4). Several studies have shown that frailty is useful to detect the heterogeneity in risk among people of the same chronological age (5, 6), including middle-aged individuals (7–9). Frailty can also predict survival, even in the absence of comorbidity (7, 10). Based on this evidence, international guidelines recommend frailty screening in order to stratify patients according to their risk of poor outcomes (11, 12). Furthermore, screening for frailty at the population level is appropriate to identify those who would benefit from a comprehensive geriatric assessment (CGA). CGA can provide the basis for a coordinated and integrated intervention to maximize the overall health status of the individual (13).

Among the available operationalizations of frailty, the so-called Frailty Index (FI), proposed by Rockwood and colleagues, is one of the most used (14). The FI measures the accumulation of health deficits (i.e., clinical signs, symptoms, diseases, and disabilities) that tend to increase with aging. The FI is obtained by calculating the ratio between the number of deficits observed in a person and the total number of deficits considered for its computation (15, 16). In several studies, FI was associated with increased mortality risk in both hospitalized (6, 17–21) and community-dwelling COVID-19 patients (22).

However, few studies have used an automated FI, based on routine collection of administrative community health data (23, 24), and none used a similar approach during the COVID-19 pandemic. In this context, electronic regional health databases (e-RHDs) represent an important resource. E-RHDs could identify people at higher risk for severe outcome or death, even when detailed information, usually collected in clinical studies, is not available.

Therefore, we undertook this study with the following objectives:

A total of 689,197 adult beneficiaries of the Lombardy Regional Health System (357,495 females, 51.9%) were positive at a nasopharyngeal swab PCR test for SARS-CoV-2 by May 20, 2021 Supplementary Figure S2. The median age was 52 years (first-third quartile, 38–66) and the Charlson Comorbidity Index (CCI) (34) was 0 for 96% of the population. The health deficits and determinants included in the e-RHD-FI are reported in Table 1 along with their prevalence. The most frequent deficit/determinant was hypertension (156,560, 22.7%) followed by low income (116,097, 16.8%) and heart disease (90,655, 13.2%). The e-RHD-FI showed a typical asymmetric distribution (Figure 1) with a median of 0.02 (first-third quartile, 0–0.05).

We developed and validated an e-RHD-FI, using information from the electronic healthcare database of the Lombardy region, Italy. The e-RHD-FI was developed according to the standardized criteria recommended by Searle et al. (16). The e-RHD-FI was validated against the c-FI and the CFS, using a set of clinical data prospectively collected in a multicentre study of patients hospitalized for COVID-19 (6). Overall, the study showed that e-RHD-FI had good performance in predicting in-hospital and 30-day mortality, risk of hospital admission, and worsening on the WHO clinical scale in COVID-19 patients.

Previous studies investigated the predictive power of the electronic FI (e-FI) for adverse outcomes in both hospitalized and community-dwelling individuals. Mak and colleagues developed a 48-item e-FI from routinely collected health records of 13,188 hospitalized older adults in Sweden, finding that a 0.03 increment in e-FI was independently associated with increased risk of in-hospital, 30-day, and 6-month mortality (35). Previously, Clegg and colleagues demonstrated that a 36-item e-FI was able to robustly predict mortality, hospitalization, and nursing home admission in a cohort of 931,541 older community-dwellers (24). Other studies, which computed an automated FI based on electronic health records, have been conducted in Canada (36), the United States (37), the United Kingdom (38), and France (39).

However, only one study used a database-derived e-FI (i.e., the Hospital Frailty Risk Score) to assess mortality in an English cohort of hospitalized older people affected by COVID-19 (40). To the best of our knowledge, this is therefore the first study to use an e-RHD-FI for predicting mortality in an adult community-dwelling population with SARS-CoV-2.

There are several findings from this study that should be highlighted. It is noteworthy that our e-RHD-FI had one of the highest performances in predicting mortality, compared to other electronic FIs. This result suggests that the variables included in our e-RHD-FI were adequately selected to represent the multiple dimensions of frailty. In fact, the e-RHD-FI included not only a list of chronic diseases usually considered in models predicting mortality but also several other conditions, known to increase susceptibility to adverse events (i.e., poor socio-economic status, sensory, cognitive, and functional impairments). Previous electronic FIs did not consider all these domains. For instance, four of the FIs included in a recent systematic review (41) did not include socioeconomic status (24, 38, 39, 42), which is a determinant of health status, whereas two other studies did not include cognitive problems or functional status (27, 37). Given its composition, we believe that our e-RHD-FI can also be proposed for non-COVID-19 patients, but future studies are required.

Another key finding of our study is that the AUC of the ROC curve for 30-day mortality was higher in the general population (AUC = 87%) than in adults hospitalized for COVID-19. In hospitalized patients, e-RHD-FI showed similar predictive power for short-term mortality to the FRACOVID study (6). The fact that frail elderly patients were less likely to be hospitalized during the pandemic, compared with robust counterparts, may partially explain our results. The e-RHD-FI was applied to the whole population, and not just to elderly patients, showing very good performance in younger adults. We also found that the ability of e-RHD-FI to predict 30-day mortality was higher than CCI, suggesting the potential benefits of assessing frailty in younger patients.

Other findings should be mentioned too. The agreement between each variable included in e-RHD-FI and c-FI was high. We found low sensitivity, but good or even excellent specificity for most of the variables included in the e-RHD-FI compared to clinical data. This was probably because electronic databases report health deficits only when they are highly pronounced. For example, in our e-RHD-FI, patients were classified as visually impaired only if they satisfied one of the two following criteria: (i) the patient benefited from an exemption (which is granted by the regional health system when contact lenses, eyeglasses, or Braille devices are required); ii) the patient had a diagnosis of visual impairment in medical records at hospital discharge. e-RHD-FI showed lower performance than c-FI in predicting in-hospital mortality (71.6% vs. 79.6%), likely because electronic databases are not as accurate as clinical databases in reporting health deficits.

The e-RHD-FI cut-offs used in our study had a low positive predictive value (14.2%) but a very high negative predictive value (99%), suggesting good performance in detecting individuals at low risk of death after a diagnosis of SARS-CoV-2. The 0.25 cut-off had optimal performance in identifying those at high risk of mortality (33%) and a low negative predictive value (96.1%), indicating that individuals with e-RHD-FI ≤ 0.25 had a 3.9% risk of 30-day mortality.

The results of this study may be useful, from a healthcare governance perspective, for managing frailty not only during but also after the pandemic. Currently, in the Italian National Health System, frail subjects are identified based only on generic characteristics, such as old age or the co-occurrence of chronic diseases. However, this approach is unable to capture the true nature of frailty and therefore is inadequate to target individuals who can truly benefit from specific interventions for frailty. The e-RHD-FI may serve this purpose, by enabling the Regional Health System to identify frail persons in a timely and simple manner. For instance, e-RHD-FI can be used to target people who will benefit most from vaccines. In this regard, surveillance systems based on predictive models adopted in restricted geographical areas have successfully prevented disease progression among people during the COVID-19 pandemic (1). The e-RHD-FI may also be used, on an individual basis, to identify individuals in need of coordinated and integrated interventions. On a regional level, the e-RHD-FI may help compare patient case mix and healthcare resources utilization across hospitals, according to frailty levels. This could have an important economic impact, given that frailty is associated with increased utilization of healthcare resources (10).

A strength of this study is that we used a large population-based healthcare database. Furthermore, we validated the e-RHD-FI against a FI based on ad hoc clinical data. We measured the accuracy of the deficits included in the e-RHD and the association between the e-RHD-FI and other well-known tools used to assess frailty (c-FI and CFS). This study also has some limitations. First, the e-RHD-FI was validated only in individuals with SARS-CoV-2 by May 20, 2021, thus excluding those who developed SARS-CoV-2 afterwards. We cannot exclude that the large fluctuations in COVID-19 infection and mortality rates observed during the pandemic, as well as the advent of vaccination, may have influenced our results. However, we expect the index to show good performance also in those who were positive for SARS-COV-2 during the third and fourth waves, as the index performance did not change significantly when we applied stratification for time of diagnosis. Furthermore, e-RHD-FI showed lower values than c-FI, possibly underestimating frailty. This might be due to the fact that it was not possible to retrieve complete information on all health deficits and determinants from administrative data (e.g., mental health). Moreover, we were unable to retrieve all the variables included in the c-FI (such as smoking status). However, the reduction in the prognostic performance was limited when comparison with c-FI was applied. Finally, the data refer only to the Lombardy region, therefore our results are not generalizable to the whole country.

To conclude, e-RHD-FI can predict mortality in community-dwelling adults with SARS-CoV-2 and can be used to target people at risk for future pandemics. Our findings also support the need to assess frailty with regional healthcare databases to implement specific approaches for frail people.

The datasets presented in this article are not readily available because they need an agreement to be accessed. Requests to access the datasets should be directed to Regione Lombardia.

The studies involving human participants were reviewed and approved by the Brianza Institutional Review Board under the number 3356-07/08/2020.

PR, CS, GB, and MGV contributed to the conception and design of the study. AZ, EF, AM, PR, GO, MGV and GB contributed to the acquisition of the data. PR, GO, and MGV were responsible for the statistical analysis. PR, CS, FB, OL, FC, AZ, EF, AM, GB, and MGV were responsible for the interpretation of the data. PR, CS, AZ, EF, AM, GB, and MGV contributed to the draft of this paper. All authors have read, revised, and approved the final manuscript.

This work was supported by a grant from the Cariplo (Cassa di Risparmio delle Province Lombarde) Foundation, Lombardia Region, Italy (The FRACOVID study, ClinicalTrials.gov Identifier: NCT04412265).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.