An observational, retrospective, multicentric study was conducted in four NHs in the Valencian Community, Spain. The study was approved by the Ethics Committee of Hospital of Sagunto (FEF-ICF-2021-01).

The recruitment period extended from January 1, 2017, to February 28, 2019. The follow-up period had to be interrupted in February 2020 to ensure that the SARS-COV-2 pandemic did not influence the risk of mortality of the study population. The inclusion criteria were as follows:

• •

  Users of the NHs Burriana, La Cañada, Puerto de Sagunto and Torrente.

• •

  Age ≥ 65 years.

• •

  Availability of a CGA including the variables considered in the IF-CSS between January 1, 2017, and February 28, 2019, to ensure a minimum follow-up period of 12 months.

The only exclusion criterion was a terminal status, defined as being in the last days of life under palliative care.

IF-CSS was organized into four dimensions (functional, mental, geriatric syndromes, and clinical), including 17 variables encompassing a total of 27 deficits (Table 1). This index is calculated by dividing accumulated patient deficits by the total possible deficits, yielding a result of 0 (no deficits) to 1 (presence of all deficits). Variables were selected according to consensus criteria, namely: association with age and health problems; representation of different domains, resistance to early saturation, and exclusion of highly prevalent or very infrequent items.17 An essential requirement was the inclusion of variables in the CGA of NHs in our environment. Pre-defined frailty intervals included: non-frail (≤0.2); mild (>0.2  ≤0.35); moderate (>0.35  ≤0.55) and advanced (>0.55) frailty.16

Each variable was scored so that the weight of each domain corresponded to that of previously validated indices.17–19 Binary variables were scored as a function of the presence or absence of the deficit. Quantitative variables derived from scales were categorized using cut-off intervals established and accepted in clinical practice (Table 1).

Statistical analysis was performed by Biostatech, Advice, Training & Innovation in Biostatistics S.L.

Considering the frailty reported in previous studies,4 a sample size of 530 patients was established, assuming a 95% confidence interval and a margin of error of 3%.

Descriptive analysis included absolute and relative frequencies for categorical variables and means and standard deviations for qualitative variables. Comparative analysis of survivors and non-survivors was carried out using Fisher's exact test for categorical variables, and chi-squared test, as appropriate for the number of patients in each category. Quantitative variables were analyzed with Welch's test and one-way ANOVA when assumptions of normality and equal variances were met; otherwise, the Mann–Whitney or Kruskal-Wallis tests were employed, according to the number of groups being compared.

The analysis of survival was carried out using the first IF-CSS value obtained for each patient during the follow-up period using Kaplan–Meier (KM) curves. Estimations for the predefined frailty intervals were performed using the log-rank test. In addition, statistical criteria were used to assess other potential categorization models that could better define frailty levels (Annex II). This analysis was also conducted for all the variables included in the construct for sex and NH where patients resided.

Given that the proportional hazards assumption for the IF-CSS variable was not met, survival time was estimated using a parametric accelerated failure time model instead of a Cox regression. An analysis of ROC curves was conducted to evaluate the prognostic performance of IF-CSS at 1 and 2 years.

A total of 535 patients were included, with a mean age of 83.62 years (SD 7.84), with 71.8% being women. The mean follow-up time was 1.99 years (SD 0.87). The population exhibited a high level of functional dependency, as shown by the median of the Barthel index (40; IQR 10–80) and the fact that 64.3% of patients had a moderate–severe functional dependency. Additionally, the mean MEC value was 16.07 (SD 11.92), with 57.2% presenting with a moderate-advanced cognitive impairment, whereas only 17.4% had no cognitive impairment. Regarding geriatric syndromes, 62.2% suffered from incontinence; 34.2% had immobility; and 33.1% were affected or at a high risk of developing pressure ulcers (PUs). In relation to diseases, a high proportion of participants had a neurological disease (75.1%), primarily dementia. Descriptive results for the study variables are detailed in Table 2, along with the number of the cases included for variables with incomplete data.

As many as 85.2% (n  =  456) of patients presented some level of frailty, which was mild in 25.8% (n  =  138); moderate in 34.4% (n =  184) and advanced in 25.0% (n  =  134). Complete data were available for all variables in over 95% of cases, and were neither excessively prevalent (>80%) nor very infrequent (<1%). The construct remains below the submaximal limit of 0.7 in 98.7% of cases, with IF-CSS values showing a right-skewed distribution (Fig. 1).

Figure 1.

IF-CSS distribution in the totality of the cohort.

The longest follow-up period was 38 months. During this period, a total of 213 deaths (39.8%) were recorded. No patients were lost to follow-up. The Kaplan–Meier estimation for the study cohort is shown in Fig. 2. The accumulated probability of survival over time was 0.541 (95%CI: 0.493–0.594).

Figure 2.

Kaplan–Meier estimation for the study cohort.

Significant differences were observed in the survival curves estimated for each frailty group (test log-rank: χ2  =  92, p  <  0.001) for the predefined frailty intervals (Fig. 3A) and other groups based on statistical criteria (Fig. 3B and C).

Figure 3.

Kaplan–Meier estimation of survival by IF-CSS level (spotted lines indicate the median for each group, whereas gray zones indicate 95% confidence intervals. The p value corresponds to the log-rank result). A) Level groups were established based on clinical criteria. B and C Level groups are based on statistical criteria.

Significant differences were noted between survivors and non-survivors in all variables except for NH and sex –not included in the construct–, polypharmacy and cardiovascular, respiratory and hepatic-gastrointestinal diseases (Table 2). Depression reached statistical significance only when assessed as antidepressant therapy, whereas the risk of falls was significant only when evaluated as risk status (Tinetti test).

The parametric accelerated failure time model for predicting survival time included the variables age, sex, NH and IF-CSS with NH variable being non-significant) The model indicates a clear effect of the variable IF-CSS, estimating a 29% reduction in survival time per 0.1 increment in IF-CSS value. Assuming all other variables remain constant, the model indicates that each additional year of age reduces the probability of survival by 3%, whereas being female increases it by 26%. Fig. 4A contains survival estimations for different IF-CSS values. As shown in Fig. 4B, the residuals conform to the distribution assumed by the parametric model. The analysis of ROC curves reveals an area under the curve at 12 and 24 months of 0.76 (95%CI: 0.70–0.82) and 0.75 (95%CI: 0.71–0.80), respectively (Annex III).

Figure 4.

A) Survival estimates by IF-CSS value and sex. Age is expressed as the mean. B) Graphical representation of the residuals of the parametric model.

IF-CSS can be easily estimated and integrated into clinical practice. The construct satisfies inclusion of variables and validation criteria,20 showing an association with age, higher mean values in women, and no saturation beyond the submaximal threshold of 0.7. IF-CSS shows a right-skewed distribution indicating that a low proportion of individuals exhibited low frailty, which can be explained by NH admission requirements.

In our cohort, the probability of survival upon follow-up completion was 54.1% (95%CI: 49.3–59,4), and 85.2% had some level of frailty. Predefined frailty intervals clearly separate survival curves by frailty level, avoiding reliance on statistically derived thresholds, which are less intuitive and more complex to apply.

As expected, a clear relationship was observed between frailty and mortality. Although this finding should be interpreted with caution, the parametric model estimates a 29% decrease in survival time per 0.1 IF-CSS increment. According to the ROC curve analysis, at 12 and 24 months the model has an intermediate prognostic value, as compared to that reported by other authors.4,15,17 These differences could be explained by variability in the clinical status of the patients at the time of frailty index estimation, the care setting, and the characteristics of the cohort.

Significant differences were observed throughout the 38-month follow-up between survivors and non-survivors in all variables except for cardiovascular, respiratory, neurological and hepatogastrointestinal diseases and polypharmacy. The high level of frailty of the cohort would minimize the impact of diseases on a comprehensive, and especially functional, status diagnosis. . Cancer and kidney failure are exceptions, likely acting as fatal triggers in advanced stages, with treatments unable to ensure resolution or prevent chronicity. Polypharmacy is a geriatric syndrome that has been associated with mortality and frailty. However, in our study, no differences were observed in this variable between survivors and non-survivors. This could result from polypharmacy being a focus of active intervention in CGA, where treatments are adjusted to patients' evolving status. CGA interventions typically lead to a reduction in the number of medications, which is consistent with the low prevalence of excessive polypharmacy (24.9%) and would interfere with the identification of an association. In our population, the clinical effects of polypharmacy –such as frailty and mortality– may relate more to the type than to the number of medications.21 In this line, the variable "delirium," which in the construct scores antipsychotic treatment, is associated with a higher risk of failure, possibly indicating that not all medications affect these individuals in the same way.

Regarding variables classified as ‘risk’ and ‘event,’ it is noteworthy that in the variable 'risk of PU', each constituent deficit shows significant independent differences, both in the risk score (Norton ≤12) and in the presence of PUs. In contrast, differences in the variable ‘risk of falls’ were only significant in risk measurement (Tinetti), probably resulting from a high proportion of individuals with moderate-advanced frailty using a wheelchair, which significantly reduces the number of falls. In the variable 'depression', the presence of symptoms (Yesavage/Cornell) is not significant, and the use of antidepressant therapy was less frequent among non-survivors. A possible explanation is that CGA care teams frequently discontinue antidepressant therapy in patients with advanced frailty.

In relation to frailty indices proposed by other authors, the Rockwood index9 has been implemented in different care settings. However, its predictive ability seems to be reduced when used with people who have a very high number of deficits, so its appplication in NH might be less appropriate.22 Frail-NH23 is the most widely used index in NHs for detecting and staging frailty; nevertheless, as it relies on frailty phenotype, it cannot be used as a continuous variable like deficit accumulation indices. We only found a frailty index based on deficit accumulation designed specifically for the NH setting.15 It should be noted that our index contemplates fewer deficits (32 vs 27), as it accounts for a reduced set of conditions, symptoms, and sensory impairments. IF-CSS is slightly below the 30 deficits recommended for the design of the construct,20 but significantly above the 10 deficits that would result in a higher risk of instability in estimates.24

The novelty of our construct lies in the use of validated scales in CGA and in distinguishing risk and occurrence of a problem as separate deficits. This new approach may enhance the consistency of estimates, as scoring is performed as a function of the severity of patient's status. Additionally, this model has a low risk of observer bias, as the different scales are applied by a team of highly experienced professionals.

Other existing indices are intended for use in ambulatory older patients4 or long term and home care settings.10 The care setting played a major role in construct design. In view of the high level of disability and cognitive impairment of NH residents, autonomy to perform instrumental activities of daily living and self-reported measures (e.g. level of pain) were excluded from IF-CSS to avoid saturation. Other deficits excluded were the level of anxiety, insomnia, and drowsiness, as these problems may be influenced by interpretation bias. Social vulnerability was also excluded as living in a NH largely solves this potential deficit.

The results of this study should be interpreted considering the nature and limitation of the study design, namely:

• •

  The first IF-CSS value for each patient was used for validation analysis, and their progress during follow-up was not taken into account.

• •

  Disease scoring is more sensitive to subjectivity, and prescriptions are assumed to be appropriate in diseases involving a drug therapy (depression and delirium).

• •

  The construct was designed for use in NH residents; therefore, their utility is restricted to this care setting.