Standard laboratory tests to identify older adults at increased risk of death

Frailty is an important problem for aging societies [1]. Increasingly, there is a sense that societies need to begin frailty screening and assessment [2], even though lack of consensus about just how to do this is acknowledged [3]. Frail older adults who become acutely ill are seen as being at particular risk, especially if exposed to the hazards of routine hospital care [4]-[6] without mitigation by specialized geriatric interventions [7].

Reflecting the lack of consensus, various scales to measure frailty are used [8]. One common approach is to quantify frailty with a frailty index (FI), based on the accumulation of health deficits [9]-[11]. These deficits can be symptoms, clinical signs, diseases, laboratory abnormalities or other measures [11]. An FI score is achieved by counting the number of deficits in an individual and dividing by the total number of deficits measured to produce a score between 0 and 1; a higher score indicates greater frailty [10],[11]. For a deficit to be included in an FI it must be shown that its prevalence increases with age, that it does not become too prevalent at some younger age and that it is associated with adverse outcomes [11]. Thus, an FI score provides a quantitative measure of health status and characterizes the risk of adverse outcomes, including death.

In clinical care, the ideal frailty screening tool would quantify frailty based on data that are collected routinely. In the hospital setting for example, admission typically is associated with a large number of blood tests and routine physical measures (for example, blood pressure) that require minimal participation by patients. Recent work on measuring frailty in mice suggests that, in combination, many such tests show minor abnormalities that, in the aggregate contribute to risk [12]. We wondered whether this might also obtain with routine physical assessment and laboratory test data that are often collected clinically. For this reason, in a re-analysis of data from the clinical examination conducted during the first wave of the Canadian Study of Health and Aging (CSHA), our objectives were: 1) to develop an FI based only on routine physical and laboratory tests (FI-LAB); 2) to validate the FI-LAB in relation to age, sex and distribution; and 3) to test its predictive value in relation to death. Here, we show that an FI based on routine laboratory tests identifies older adults at an increased risk of death.

Of the 1,375 people with both clinical examinations and laboratory data, complete data were available on 1,013, of whom vital status was known for 986 (97.3%; Additional file 1: Figure S1). Selected demographic and clinical characteristics of the study population, subdivided by grades of frailty for both the FI-CSHA and the FI-LAB, are illustrated in Table 2. The mean frailty scores increased with age for both frailty measures. Mean FI-CSHA scores increased from 0.07 ± 0.02 in the least frail group to 0.50 ± 0.05 in the frailest group (Table 2). Similar results were seen when the FI-LAB scores were used to stratify frailty. The average FI-LAB values increased from 0.08 ± 0.02 in the group with the lowest scores to 0.50 ± 0.04 in the group with the highest frailty (Table 2). Of note, the proportion of women with low FI scores was much higher when frailty was stratified by FI-LAB scores compared to the FI-CSHA. The mean combined FI scores also increased from 0.08 ± 0.01 in the group with the lowest scores to 0.50 ± 0.04 in the group with the highest frailty scores (Table 2). The characteristics of the 372 excluded cases (mean (±SD) FI-CSHA = 0.26 ± 0.12; mean age = 81.9 ± 7.9 years; 64.4% women) were similar to those of the 1,013 included cases.

To compare the distribution of FI scores for the three different FI instruments used in this study, frequency distributions for each were plotted. Figure 1A shows a frequency distribution of the FI-CSHA scores obtained for the cohort investigated in this study. The distribution was slightly skewed to the left, with a mean of 0.25 ± 0.11 (±SD) and a median of 0.24. The minimum FI-CSHA score observed was 0.02 while the maximum was 0.72 (Figure 1A), consistent with the idea that there is a sub-maximal limit to frailty near 0.7. The frequency distribution for the FI-LAB scores is shown in Figure 1B. This distribution had a mean of 0.27 ± 0.12 (±SD) and a median of 0.27. The minimal and maximal FI-LAB scores were 0.05 and 0.63, respectively (Figure 1B). Figure 1C shows that the frequency distribution for the combined FI scores was similar to the distribution of the two parent index scores. This distribution was slightly skewed to the left with a mean of 0.26 ± 0.09 (±SD), a median of 0.25 and minimal and maximal scores of 0.06 and 0.59, respectively (Figure 1C).

The log of each FI score increased linearly with age (data not shown). The r² values were 0.57 for the FI-CSHA, 0.62 for the FI-LAB and 0.69 for the combined FI. Regression lines fitted through the data had slopes of 0.015, 0.012 and 0.013 for the FI-CSHA, the FI-LAB and the combined FI, respectively. To determine whether the FI-CSHA scores and the FI-LAB scores were linearly related, the FI-CSHA scores were plotted as a function of the FI-LAB scores (Figure 2). The average FI-LAB scores increased as the FI-CSHA increased; this relationship was a very good fit to a straight line (r² = 0.81).

Mortality rates generally increased as the frailty scores rose, although this effect was more marked with the FI-LAB than with the FI-CSHA scores (Table 2). Mortality also increased significantly as the frailty scores increased (Table 2, Figure 3). Note that in an age and sex adjusted model, each contributed independently: the odds ratio for the FI-LAB =1.03, 95% CI =1.01, 1.04, versus FI-CSHA = 1.04, 95%CI = 1.02, 1.05. The combined FI showed the clearest separation of groups by grades of frailty (Figure 3), and was associated with the highest hazard rates in age and sex adjusted models (Table 3). The impact on the discriminative ability of combining both FIs was modest: the area under the receiver operating characteristic (ROC) curve was 0.71 for the FI-CSHA, 0.72 for the FI-LAB and 0.74 for the combined FI (Additional file 1: Figure S2). Nonetheless, together these data show that all three FI scores identified older adults at increased risk of death.

We investigated the properties of an FI made up of information from widely used laboratory tests. That FI (the FI-LAB) had properties similar to other FIs, including the FI-CSHA. The latter consists of up to 38 items from the CSHA clinical examination and corresponds to most of the items used in a Comprehensive Geriatric Assessment, which can be summarized as an FI [5],[6],[16],[17]. Even so, the people classified as at risk by each method differed in interesting ways. Although the distributions of the two FIs were similar, as were their mean ages, fewer people had the lowest FI scores in both FIs. For example, only 15 people had a combined FI score less than 0.10, compared with 78 for the FI-CSHA and 56 for the FI-LAB (Table 2). Of these 15, only 1 died (Figure 3), the lowest mortality rate of the three FIs (combined, FI-LAB and FI-CSHA). The analogous case holds for people with the highest scores in each of the FI-LAB (n = 57) and FI-CSHA (n = 39). Only 17 individuals had the highest scores in both and of these 15 died (Table 2). This 88% five-year mortality was the highest for any FI category. These large differences in mortality in relation to the combined FI categories were observed even though the increase in the area under the curves (AUCs) was modest, suggesting that the middle classifications could be more finely graded. Even so, we have resisted finer grades of the combined FI, on the grounds that, from a clinical point of view, knowing the highest and the lowest risks is most important: people of intermediate risk represent variations on the usual case, and typically receive usual management without useful precision in prognosis. In any case, testing how much information comes from the nature of the added items, and how much comes from their number is best addressed in a different datatset.

Additional questions remain, however, such as whether the smaller number of people with the fewest things, or most things, wrong reflects a specific increase in information from laboratory data, or whether that might be shown simply by increasing the number of items considered in an FI. Here, the distribution of the items in each FI suggests some comparability (and is reassuring in relation to combining items), but their independent contribution in a multivariable model suggests that they are offering independent information. The latter suggests that subclinical information, or at least information more precisely detected with laboratory tests, could offer additional insights. As detailed elsewhere [21],[22], frailty that is macroscopically detectable represents the build-up of subcellular, tissue and organ deficits, being damage at those levels that has gone either unremoved or unrepaired. The lethality of any clinical/macroscopic deficits on a background of subclinical/microscopic deficits is suggested by the combined FI hazard rate being the highest, and by the notably diminished survival of the frailest group (FI >0.45) in the combined FI (Figure 3, Panel C, compare with Panels A and B).

Our data must be interpreted with caution. Our sample, although population-based, is not representative. Given that it was drawn from the clinical examination database, it is older and contains proportionally more institutionalized people than the population from which it was drawn. Even within the clinical sample, developing an FI with items that were relevant for both community-dwelling and institutionalized people was a challenge: for example, no one in institutional care is independent in any instrumental activities of daily living (ADLs), and few community-dwelling people have fecal incontinence or behavioral problems. In consequence, meeting all criteria for creating an FI-CSHA with sufficient non-missing data meant that only 1,013 of 1,375 people with laboratory test information could be used. Even so, relaxing the criterion to allow up to 18 (of 38) variables to be missing (compared with the usual requirement for not calculating an FI for anyone for whom more than 20% of the items are missing) added another 158 subjects, without changing the properties of the resulting FI. Similarly, the mean age of the 362 people for whom we did not calculate an FI-LAB was 81.9 years (64.4% female) compared to 81.1 years (61.6% female) for those for whom we did calculate an FI-LAB. Likewise, the mean FI-CSHA score was 0.26 when all 1,375 subjects were used compared to 0.25 when only the 1,013 people investigated in this paper were considered. Still, understanding whether laboratory data will add value in more representative samples, or in other clinical samples, requires cross-validation in other datasets. In other settings, different or additional laboratory tests might be used reflecting local differences in relation to congenital and acquired disease, or those associated with lifestyle or environment (for example, thalassemia, histoplasmosis, alcoholism, air pollution). More recent reports might also change which tests were selected, for example, substituting alanine transaminase for aspartate transaminase [23]-[25].

Here, we tested the predictive validity of the frailty measures in relation to mortality. With each version of the FI, higher FI scores were associated with greater mortality, verified in a multivariable model that included age. Although death has the advantage of being a dichotomous, unambiguous and relevant example of an adverse outcome, not everyone dies a frail death. In consequence, it is important to distinguish between the exercise of predicting mortality per se and using it to validate the notion that more deficits are associated with greater risk. For example, were mortality prediction to be all that motivated our inquiry, then the FI would include chronological age, as evidenced by its persisting significance in each of the Cox proportional hazards models (Table 3). In the present context, this would, of course, be perverse: the goal of defining frailty is to address why, even though age is highly associated with the risk of adverse health outcomes (including death), not everyone of the same age has the same risk. The suggestion of the FI - that people with the most things wrong are at the highest risk - also has the advantage of being parsimonious, and of being sensible on its face, which itself is another form of validation [26].

The addition of laboratory test data also is of interest in understanding the associations between specific test abnormalities and frailty, as is commonly undertaken in inquiries about putative frailty mechanisms. For example, some groups have evaluated individual or even small numbers of laboratory tests [27]-[30]. A burgeoning literature on biomarkers, typically motivated by discrete mechanistic hypotheses, often considers such tests individually [31]-[34]. Our data suggest that any such results need to be interpreted in the context of the overall health state of the organism, if a general claim about frailty is to be made. Consider that the mean value of the FI score is typically closely related to age - or, as has been argued elsewhere - is a measure of biological age [10],[35],[36]. Aging is associated with a very large number of cellular and tissue mechanisms [37]. Finding associations with any single test abnormality can be helpful, but understanding where this fits in relation to other test abnormalities is an important step in aiming to gain a mechanistic understanding and in understanding systems effects. Our data suggest that, particularly for those sorts of inquiries, looking at the contribution of any single health deficit in isolation will be pragmatically difficult and theoretically dubious. A similar argument obtains in relation to understanding age-related mechanisms. Support for the latter comes from a recent study which shows that later life changes in myocyte structure and function are more closely tied to an FI than to chronological age [12]. As argued elsewhere, extending work on frailty to animal models can allow for exploration of mechanisms of both frailty and, more broadly, aging itself [38]. Finally, the observation here that more women had lower FI scores using the FI-LAB than using the FI-CSHA is of interest. Of note, either way women had lower mortality than did men for any level of either FI. The more conservative estimate of frailty status observed with the FI-LAB might be an explanation for the so-called male-female mortality-morbidity paradox [39]. This intriguing observation needs to be pursued further.

The results of this study demonstrate that an FI constructed from routinely collected laboratory and clinical data identifies older adults at increased risk of death. A large number of additional inquiries are suggested by these current findings. Beyond replication, and as a probe for understanding mechanisms, the feasibility and utility of adding a large number of items to an FI using commonly evaluated laboratory tests might importantly advance routine frailty assessment, especially when these test results are used in conjunction with other relevant items from electronic medical records. These considerations are motivating additional inquiries by our group. In particular, further evaluation in clinical settings of adding routinely collected laboratory data to an FI is warranted.

SEH and KR conceived of the study. SEH, MRHR, AM and KR participated in the design of the study and performed the statistical analysis. SEH and KR helped to draft the manuscript. All authors read and approved the final manuscript.