Frail-VIG index: a concise frailty evaluation tool for rapid geriatric assessment

Aims

Based on the multidimensional approach of the CGA and inspired by the simplicity and potential universal applicability of the Rapid Geriatric Assessment (RGA) [34], together with the known value of the FI methodology, we have developed a new FI called the “Frail-VIG index” (VIG is the Spanish/Catalan abbreviation for CGA). In the present observational study, we prospectively applied the Frail-VIG index to 590 patients admitted to the Acute Geriatric Unit (AGU) at the University Hospital of Vic (Barcelona, Spain) to determine the instrument’s capacity to predict mortality at 24 months.

Study design

This was a prospective, observational, longitudinal study.

Context and participants

The study was conducted at the University Hospital of Vic (Barcelona; Spain), a 200-bed acute care hospital covering a population area of 156,000 inhabitants. All patients admitted to the AGU during the year 2014 were included. Admission criteria to the AGU were: 1) age ≥ 85 years; or 2) cognitive decline; or 3) the presence of advanced chronic conditions identified by the NECPAL (NECesidades PALiativas, in Spanish; Palliative Care Needs, in English) test [35], a validated tool for the early diagnosis of the need for palliative care among individuals with limited life expectancy. No exclusion criteria were defined.

Design and evaluation of the frail-VIG index

To design the most pragmatic instrument possible, the Frail-VIG index was constructed using only variables recorded during the usual clinical evaluation process—as opposed to variables collected as part of an epidemiological study, which is how most other FIs were developed [12, 15, 18, 36, 37]. In our department, the original aim was to systematically perform a CGA for all patients; to achieve this, all health care professionals in the department agreed by consensus to create a checklist that included 40 multidimensional variables. Subsequently, given the need for a quantifiable RGA for decision-making at our acute care hospital [33], a proposal was made to reduce the number of variables, prioritizing those variables with the strongest prognostic capacity for mortality [38].

In the development of the Frail-VIG index, rather than comparing the index to a “gold standard” instrument, we sought to replicate the characteristics of a previously validated instrument [3, 4, 12, 14, 15] and then we evaluated its predictive ability [39]. Our development process focused on two main areas: 1) variable selection and construct development, and 2) demonstration of the predictive validity for mortality. The design of the Frail-VIG index has been previously described [40].

Variables and data sources

In the calculation of the Frail-VIG, binary variables are scored as “0 points” (no deficit) or “1 point” (presence of the deficit), with the exception of advanced chronic illness (defined according to the NECPAL criteria; see Additional file 2), which was scored as 2 points. For ordinal variables, the Frail-VIG uses the well-accepted cut-off points used in routine clinical practice for the Barthel Index (BI), which is used to assess basic ADLs, categorized according to the Saha criteria [46] and Reisberg’s Global Deterioration Scale (GDS), which assesses cognitive impairment, categorizing patients into one of 3 groups [47].

To prevent intercurrent illnesses (that is, the conditions that led to hospitalization) from influencing the Frail-VIG score, the index score identifies only the patient’s baseline condition (that is, ≥ 1 month before hospitalization and/or prior to onset of the clinical process leading to hospital admission). This information is collected on the patient’s first day in the hospital by interview (patient and/or family) to obtain a full medical history and/or by checking the medical record to determine current prescriptions.

Statistical methods

Descriptive statistics of the variables were calculated using the SPSS software program (IBM; Chicago, IL; USA). Statistical significance (p < 0.05) with 95% confidence intervals was obtained by comparing mean values (for quantitative variables) and percentages (for qualitative variables).

Survival estimates were obtained for the 24-month follow-up. The survival analysis was performed using the Survival, pROC, and RMS packages from the R project (https://www.r-project.org). Survival curves were computed according to the Frail-VIG index scores, grouping these according to predefined criteria before the analysis into seven interval ranges (0–0.15, 0.16–0.25, 0.26–0.35, 0.36–0.45, 0.46–0.55, 0.56–0.65, and 0.66–1), and then the survival curves were compared via the log-rank test. The ROC curves were analysed to determine the predictive capacity of the Frail-VIG index.

Cohort description and analysis of the variables

Variable selection and results of construct evaluation

The deficits included in the Frail-VIG index are those commonly associated with age and adverse health outcomes and these deficits were carefully selected to represent the main domains to be assessed (see Additional file 1).

The prevalence rate for each variable was < 80% (except for polypharmacy, which was exactly 80%) and none could be considered rare (< 1%). Although the prevalence for some variables—for example, polypharmacy—was high, none of these variables are universal, even in 85-year old patients [48]. The overall construct validation process [4] shows that the distribution of variables by domain (as a percentage of the overall index) was similar to other validated FIs (see Additional file 1). Despite the high mean age of this population cohort and the large percentage of individuals presenting with advanced chronic conditions, the submaximal limit (0.7) for FI scores was maintained (that is, the Frail-VIG index was < 0.7 in 99.3% of patients). The distribution of scores on the Frail-VIG index tended towards asymmetry (asymmetry coefficient, 0.37) (Fig. 1).

Mortality at 24 months

The cohort was followed for 24 months or until death, whichever came first. At 24 months, the mortality rate was 57.3% (n = 338). None of the patients were lost to follow up. Table 2 shows differences in mortality rates for each individual study variable. The mortality rate increased progressively in line with higher Frail-VIG scores (Fig. 2), especially at the 12 month follow up (Fig. 2; panel A).

We evaluated the correlation between mortality and the Frail-VIG index by means of the log-rank test (Fig. 3), comparing the seven different survival curves based on the score intervals described in the statistical analysis section above. Highly significant differences were observed between the various range intervals (X² = 433.4, p < 0.0001). Table 3 shows the number of individuals in each frailty index category who had died at 24 months of follow-up.

Frail-VIG index category		0.1 (0–0.15)	0.2 (0.16–0.25)	0.3 (0.26–0.35)	0.4 (0.36–0.45)	0.5 (0.46–0.55)	0.6 (0.56–0.65)	0.7 (0.66–1)	Total
12 months Follow-up	Alive	22	72	54	113	50	5	0	316
	Deceased	0	1	7	48	73	115	30	274
	Deceased (%)	0	1.4	11.4	45.3	70.7	97.5	100	46.4
24 months Follow-up	Alive	19	61	45	88	36	3	0	252
	Deceased	3	12	16	73	87	117	30	338
	Deceased (%)	13.6	16.4	26.2	45.3	70.7	97.5	100	57.3
Total		22	73	61	161	123	120	30	590

On the ROC curve analysis, the area under the curve (AUC) values at 3, 6, 9, 12 and 24 months were, respectively, as follows: 0.87 (range, 0.84–0.90), 0.88 (0.85–0.91), 0.89 (0.87–0.92), 0.9 (0.88–0.92), and 0.85 (0.82–0.88) (Additional file 3). The optimal cut-off point was 0.46 in all cases, revealing a better performance (as measured by the Youden index) at 12 months (0.62), with a sensitivity and specificity of 0.80 and 0.83, respectively (Additional file 4).

According to the participating clinicians, administration time for the Frail-VIG index ranged from 5 to 10 min.

Key results

At the 24 month follow-up, we found statistically significant differences (p < 0.05) between the two groups (living vs. deceased) for most of the Frail-VIG variables, a finding that confirms the index’s discriminative capacity. The only variables that were not statistically significant were 1) social vulnerability—likely due to the limited number of patients with social vulnerability (19 patients; 3.2%), 2) emotional status (although there were significant differences [p 0.016] at the 12-month follow up), and 3) falls (probably because we did not differentiate between severe and non-severe falls).

These results, which met the previously-specified internal validation criteria, support the validity of the Frail-VIG index [16, 44]. All criteria were fully met, both in terms of the variables included and with regard to the construct’s validation criteria.

The strong correlation between the Frail-VIG index score and mortality was consistent with published data [14, 17]. The 12- and 24-month AUC were 0.90 and 0.85, respectively, demonstrating that the Frail-VIG index offers a better prognostic capacity than other previously published FIs (whose AUCs range from 0.6 to 0.8) [4, 14, 49, 50]. A study conducted by Pilotto et al. evaluated several different FIs [51] in a group of hospitalized patients (20 Italian geriatric units; n = 2033), reporting a 12-month AUC of 0.69 for the FI-SOF (Frailty Index-Study of Osteoporotic Fractures) [52], 0.73 for the CSHA-FI ⁷⁰ , 0.73 for the FI-CGA, and 0.75 for the MPI. A study by Ritt et al. involving 307 hospitalized inpatients in an AGU in Erlangen (Germany) compared the predictive capacity of 3 different FIs based on CGA, finding an AUC at 6 months that ranged from 0.77 to 0.84 [53].

Although the excellent predictive capacity of the Frail-VIG index may seem surprising, we believe that the following factors may have positively influenced our results: (1) The characteristics of the instrument itself, which incorporates only those variables that are routinely used in conventional CGAs. Moreover, most of these variables are good predictors of mortality, as evidenced by the statistically significant differences between the two groups (alive vs. deceased) for practically all of the individual variables included in the index (Table 2). (2) There were no missing data from the baseline data due to the systematic data collection processes during routine clinical practice of the health care team. Similarly, no data were missing during follow up thanks to the reliability of the “Shared Medical Record” system in Catalonia (HC3). (3) The instrument was administered by a small group of professionals with extensive expertise in geriatric assessment. Although the questionnaire is both short and simple (containing only basic questions with dichotomous response options), we do not know if administration of this instrument by untrained staff or in other settings could lead to different results; nor do we have any data on possible interobserver variability. (4) We cannot rule out the possibility of selection bias caused by the AGU admission criteria, which could have led us to select a population with high mortality rates. (5) The intercurrent process that led to hospital admission may have affected the final mortality results, potentially increasing specifically the mortality rate in those patients who were the most frail prior to admission.

When we compared the results at the 12 and 24 month follow up periods, we found that the prognostic accuracy of the index tends to decrease over time (Fig. 3, Table 3 and Additional files 3 and 4). The most plausible explanation for this finding is the dynamic nature of frailty [54]: the Frail-VIG index score for each patient reflects a static vision of reality; that is, it indicates the patient’s status only at the time of data collection. However, given the patient profile of this cohort (elderly patients with high multimorbidity rates), it is logical to expect that these patients will continue to accumulate deficits during the follow up period, thus increasing their Frail-VIG index score; this would also explain 24 month mortality in individuals whose initial Frail-VIG score (< 0.25) did not indicate the presence of frailty.

Administration of the Frail-VIG index does not preclude the use of the CGA for systematic multidimensional assessment [55], but instead offers the advantage of a RGA: only 5–10 min of time are needed to complete the Frail-VIG; by contrast, the CSHA-FI⁷⁰ [12] requires around 20–30 min [21] and the FI-CGA [23] about 25 min [56].

Utility of the frail-VIG index

Although more studies will be needed to confirm the benefits of using FIs to assess geriatric patients, we believe the Frail-VIG index to be a valuable tool:

1. (1)

   The index can accurately discriminate between frail and non-frail patients. The fact that more than 80% of the patients admitted to the AGU—a unit specifically designed for frail patients—are, in fact, frail, supports the appropriateness of the unit’s admission criteria.

    
2. (2)

   The Frail-VIG index score assesses the degree of frailty as a continuous variable. This can help clinicians make an accurate situational diagnosis [31]: the ability to determine the patient’s approximate biological age (degree of frailty) relative to his/her chronological age, can be particularly relevant when selecting or adjusting the treatment to develop an appropriate therapeutic plan. For instance, identifying advanced, progressive frailty (after potentially reversible causes have been ruled out), helps in analysing the risk/benefit of aggressive interventions. Conversely, assessing the degree of frailty can also be useful in decision-making for patients at risk of undertreatment who, though having fair reserve, could be denied additional diagnostic efforts or specific therapeutic interventions due to age. Frailty assessment using FIs allow us to reduce uncertainty by providing better understanding on the patient’s overall situation, helping to improve the dialogue between the patient, family and health care professionals in regards to expectations and objectives in the shared decision-making process (Fig. 4).

    
3. (3)

   Related to the previous point, quantification of the degree of frailty by means of the Frail-VIG index could facilitate the creation of a common reference for communication with health care professionals in other specialities. In addition, this could potentially allow clinicians to monitor the results of any interventions. Finally, quantification could be highly valuable in patient care and for research purposes [18].

    

Study limitations and next steps

The data described here should be interpreted with caution given that our cohort is not representative of the general population. An important limitation of the present study is that the Frail-VIG index has not been validated by comparing it to other instruments in the same cohort, nor compared to other variables (for example, institutionalization or use of resources). Additional studies are required to assess the reliability and reproducibility of this index in other settings and populations. Some study variables were obtained by reviewing the patient’s medications without confirming the suitability of the prescription and/or dose for the patient’s condition; this approach to obtaining certain variables may have biased the study outcomes. Assessment of certain variables (for example, social needs) needs to be improved to increase the reliability and accuracy of the tool. Finally, although the professionals agreed about the time needed to administer the instrument (that is, more than 5 min but less than 10 min), we did not systematically measure (using a timer) these data.

Finally, the next steps in the development of the Frail-VIG index are the following: (1) validate the instrument in an independent cohort, in other settings and distinct populations; (2) demonstrate its applicability and utility for professionals in routine clinical practice; and (3) as with most FIs, the true challenge is to demonstrate whether the routine, systematic use of this tool actually improves the health outcomes of the patients in the “triple aim” framework [57].