The Nursing Home Frailty Scale: An Efficient Approach to Assessing Frailty in Long-term Care

Frailty is a complex, multidimensional state that consists of a critical combination of impairments.^1,2 It has been operationalized by measures such as weight loss, weakness, self-reported exhaustion, decreased gait speed, decreased physical activity, dependence with dressing, restricted walking, and incontinence.^3,4 Frailty thus incorporates multiple functional and health domains and spans a continuum from a limited number of persistent problems to complete disability.⁵ One study found that 77% of older adults who were hospitalized as a result of a collapse or fall and who had a discharge diagnosis of an “ill-defined condition” were characterized as having frailty.⁶ Declining health status, self-perceived physical limitations, and cognitive decline are among the negative outcomes that have been associated with frailty.⁷ Conversely, older adults who demonstrate a good sense of personal control, psychological security, and autonomy have been shown to have a decreased chance of becoming frail.⁸

Frailty is common among residents of long-term care facilities (LTCFs). It is a complex health and functional state that over time can translate into higher rates of falls, weight loss, functional decline, hospitalization, and death.^1,9-11 Thus, the development and implementation of a universal, population-based scale to identify LTCF residents who are frail contributes to greater understanding of their current and future life events, whether they be improvements in function or nutritional status or continued decline resulting in hospitalization or death.¹²

Historically, early efforts to measure frailty began with a focus on a short list of relevant clinical conditions. Fried and colleagues⁸ provided a frailty phenotype with the criteria of unintentional weight loss, self-reported exhaustion, weakness as measured by grip strength, slow walking speed, and a low level of physical activity. The presence of 3 or more of these criteria indicated frailty. Using this measurement approach, few items are included in the evaluation, and there are a limited but quite focused set of options for designating whether or not a person is frail. Deficit accumulation is another approach to the assessment of frailty, wherein a frailty index is calculated based on a person’s number of symptoms, signs, diseases, and other deficits.^4,13 In a review of 27 population-based studies, Bouillon and colleagues¹⁴ found that the range of domains used to measure frailty included physical function, disability, disease, sensory impairment, cognition, nutrition, mood, and social support.

In examining the concept of frailty for this article, we followed the deficit accumulation approach, integrating functional, cognitive, clinical, and treatment indicators to best reflect the status of the LTCF resident. Using a typical frailty scale, the higher the score, the more impairments are present, and the greater likelihood that an elevated frailty scale score will have a bearing on a person’s subsequent life course. At the same time, we focused on predictive elements that were not likely to be subject to rapid reversal—items such as fever, pneumonia, bleeding, and fall with injury—so as to ground the new frailty scale on more-stable items.

In what follows, we discuss this new comprehensive frailty scale for LTCF residents. The interRAI Nursing Home (NH) Frailty Scale is based on data from a national cohort of US NH residents collected as part of the Centers for Medicare and Medicaid Services (CMS) Minimum Data Set (MDS) NH assessment instrument, along with clinical judgment about residents’ proximity to death. We discuss how the scale might help to understand residents’ status in areas that were not directly included in the set of items that defined the scale—specifically, measures of walking, bladder control, and memory. We also evaluate how recovery, as measured on the scale might affect outcome scores for the same measures, namely walking, bladder control, and memory. A resident’s frailty scale score can improve,^15,16 and we examine whether such a recovery as measured on the scale extends into other functional and clinical areas.¹⁷

Data Source

The study cohort consisted of US NH residents for the year 2012, all of whom were assessed with MDS Version 3.0. As a federally mandated process for all residents in certified NHs, the MDS obtains data reflecting a comprehensive geriatric assessment. The MDS is mandated to occur once every 90 days, however, there is a 30-day period allowed to complete the assessment. The baseline sample at initial assessment comprised 1,097,189 persons, while the 90-day follow-up assessment group comprised 1,007,450 persons. In completing the death measure (the key dependent measure of overall frailty), we reviewed all completed MDS assessments for the resident during the ensuing 120-day period.

Identification of Scale Items 

Following the recommendations of Searle and colleagues,¹⁸ the scale’s construction is focused on independent variables that have been shown to be related to a key dependent measure of overall frailty—in this case, death in the NH in the 120-day period following the baseline assessment. Using a secondary analysis approach, the NH Frailty Scale was developed from a subset of items derived from the interRAI Long-Term Care Facilities Assessment System and the MDS Version 3.0 NH assessment instrument. Both provide a basis for assessing a resident’s risk of decline in key areas that typically impact LTCF residents.

The founders of interRAI, an international collaborative network of researchers and practitioners from over 35 countries, were contracted by the Health Care Financing Administration (now the CMS) to develop a standardized data set as part of the federal Nursing Home Reform Act. In 1990-1991, the original version of MDS was implemented in 17,000 US NHs. The MDS since has been revised twice, and MDS Version 3.0 and the interRAI LTCF Assessment System are the offspring of the 2 original standardized versions of the MDS assessment created by interRAI.¹⁹

Independent Variables

A total of 113 independent variables, organized within 5 domains, were considered for possible inclusion in the NH Frailty Scale. Included were measures of function, cognition/communication, clinical status, medical diagnosis, and treatment. The function set included a full panel of measures of activities of daily living. The cognition/communication items included memory, decision-making, dementia, hearing, and delirium. Clinical status items included an array of more-stable indicators of resident status such as weight loss, pain, bone health, urinary tract infection, respiratory status, incontinence, dizziness, edema, oral problems, swallowing difficulty, and skin conditions. The medical diagnosis subset included measures such as cancer, kidney failure, heart failure, musculoskeletal disease, neurological disease, malnutrition, and diabetes. Treatment categories focused on wound care, respiratory care, intravenous (IV) line care, catheter use, transfusions, and oxygen therapy.

Dependent Variables

The dependent variable used in the logistic equation to bring together the independent variables in the frailty scale was death within 120 days of the baseline assessment. (While the MDS mandates that each resident is assessed every 90 days, it allows a 30-day period to complete the assessment). This measure represents the known terminal status of the resident during the follow-up period as recorded on the MDS Version 3.0 assessment tool. If a resident left the NH and then died, we did not have that information. In an attempt to account for residents’ leaving followed by death, we reviewed how a measure that referenced either death or hospital discharge without a return to the facility related to the NH Frailty Scale. We also assessed how well the NH Frailty Scale related to yet another relevant death-like measure that was not considered in creating the scale: a clinical judgment at the time of the baseline assessment of whether the resident’s life expectancy was less than 6 months.

We next looked at how the scale related to a short list of measures that were not included in the final scale construction. These measures come from the 90-day follow-up assessment include walking, bladder incontinence, and memory. Two forms were created for each measure: one based on cross-sectional dependence (ie, status of the problem) at follow-up and the second based on decline between baseline and within 120 days of the baseline assessment. In these analyses, we looked at how baseline NH Frailty Scale score and recovery over time on the NH Frailty Scale were associated with the residents’ scores for these 3 dependent measures. For example, is a dependent status for bladder incontinence in some way different for those whose frailty risk status improved? In the analyses, we assessed how the residents’ status on this subset of dependent measures within 120 days of the baseline assessment was altered depending on whether there was an improvement (recovery) of 1, 2, or more points on the NH Frailty Scale.

Analysis

Data were provided pursuant to an agreement with CMS to assess the multidimensionality of residents’ status in NHs, as the MDS is a comprehensive assessment that includes the areas of function, cognition/communication, clinical status, medical diagnosis, and treatment. The analyses were approved by the institutional review board of the Institute for Aging Research at Hebrew SeniorLife and were completed using SPSS Statistics 20 statistical analysis software.

The initial analytic step identified those independent variables from the large pool of items reviewed in which the person’s status on the measure was related to probability of death. A “best” dichotomous form was created for each measure, and the “death” univariate odds ratio (OR) was calculated. Variables with a univariate OR of 1.3 or higher (indicative of at least a moderate relationship to death) were considered for inclusion in the frailty scale.

Next, the dichotomous measures were assigned to 1 of 5 domain types: function, cognition/communication, clinical status, medical diagnosis, and treatment. Within each domain, the variables were subjected to logistic regression analysis to identify those that made a unique contribution to the outcome, ie, death or discharge to hospital without return to the NH.

Finally, the measures identified in each of the 5 domains were put through a final logistic regression analysis with death. The items arising from this step were summed to create the NH Frailty Scale. The scale was then compared with the initial death-dependent measure and the 2 other death-related measures, namely, discharge to hospital without return to the NH and clinical judgment that the resident was close to death.

Considering the potential utility of the NH Frailty Scale as a clinical decision-making tool, we also examined whether follow-up prevalence or change estimates for the 3 selected dependent measures (walking, bladder incontinence, and memory) tracked with a resident’s score on the frailty scale. More specifically, first, did increased dependence correspond with increasing scores on the NH Frailty Scale?  Second, was there a positive relationship between improvement on the NH Frailty Scale score and lower rates of decline on each of the 3 measures?

 In the 120-day period after the baseline MDS assessment, 6.7% of residents in US NHs died (baseline sample = 1,097,189 assessments). Domain-specific logistic regression analysis identified the following sets of measures:

• The function domain included 6 items: walking in room, locomotion on unit, locomotion off unit, dressing, eating, and bathing.

• The cognition/communication domain included 6 items: understanding others, making oneself understood, inattention, altered consciousness, mental status change, and delirium.

• The clinical status domain included 12 items: urinary tract infection, 2 measures of shortness of breath (with exertion and sitting at rest), unhealed pressure ulcer, foot infection, foot lesion, weight loss, venous ulcer, swallowing problem, bladder incontinence, bowel incontinence, and depression.

• The medical diagnosis domain included 4 items: cancer, renal insufficiency, heart failure, and malnutrition.

• The treatment domain included 6 items: indwelling catheter, application of dressings, oxygen therapy, transfusions, IV medications, and IV feeding.

Using these 34 items as input, the final logistic equation, which assessed all of the above items, identified 21 dichotomous measures for inclusion in the NH Frailty Scale. The accompanying Table 1 displays descriptive statistics for each of these measures, including the percentage of residents who had the problem, the measure’s univariate OR with the death outcome, the measure’s multivariate OR with the death outcome, and for the mutable items (ie, those in the function, cognition/communication, and clinical status domains), the percentage of residents who had the problem at baseline but did not have it at the 90-day follow-up.

Article continues below

Baseline problem rates were highest for items in the function domain and lowest for items in the clinical status domain. For example, in the function domain, 26.4% of the residents received extensive help for eating, 72.3% for dressing, and 58.6% for locomotion off the unit. All but 1 item in the clinical status domain had a prevalence rate of less than 10%; the average was 5.7%.  Clinical status items with rates at or lower than the average were shortness of breath, sitting (2.7%); unhealed pressure ulcer (5.7%); venous ulcer (1.4%); and swallowing problem (5.0%). Because the cohort was more than 1 million residents at baseline, even at these low rates, significant numbers of residents were affected by these problems, thus the frailty scale included these items.

Table 1 also shows the rate of recovery for items in the 3 mutable independent variable domains: function, cognition/communication, and clinical status. The function domain, which included the highest proportion of residents of the 3 problem domains, had the lowest rate of recovery among residents with the problem—an average of 6.7%. Cognition/communication sat in the middle at an average recovery rate of 28.4%, and clinical status had the highest average recovery rate at 39.2%. Items with a recovery rate of 50% or more in the cognition/communication domain included only mental status change (60.5%), while in the clinical status domain, only weight loss (58.7%) met this criterion. Of these items with a recovery rate of 50% or more at follow-up, weight loss had the highest baseline prevalence at 6.8%.

Frailty Scale Count Distribution

Figure 1 displays the baseline distribution of the NH Frailty Scale, derived from the sum of the dichotomies for the 21 independent variables in Table 1. The presence of a problem was assigned a score of 1. Here, individuals with a score of 14 or higher were assigned a score of 14. Even though the problem count could have been as high as 21, only 0.1% of persons in the US NH sample had a score of 14 or higher. The mean score was 4.45 (standard deviation, 2.58), the median was 4. The mean and the median scores were based upon the distribution of scores from 0 to 14. The KR-20, reliability coefficient, was 0.69, representing good internal consistency among the scale items.

The scale scores thus extended from 0 to 14. There was a progressive decline in the percentage of persons with higher scale scores beyond the mean and median. Approximately 12.3% of the sample population had frailty scores of 8 to 14. In terms of stability over time, the Pearson correlation coefficient between the score at baseline and the score at the 90-day follow-up was 0.88, reflecting minimal change among the residents with higher frailty scores. Although not included in Figure 1, over this time period, 24.7% of all residents in US NHs moved from a higher to lower score on the NH Frailty Scale—16.9% improved by 1 point, while 7.8% improved by 2 or more points.

Figure 2 displays information relative to the death status of the residents across the categories of the NH Frailty Scale—either as a total death rate, or as a death rate plus an estimate of those residents who were discharged to a hospital and never returned to the NH, or as the proportion of residents who at baseline assessment had a prognosis of living less than 6 months, representing a clinical judgment. In each instance, the rates rose across the categories of the NH Frailty Scale.

The proportion who died remained below 10% for NH Frailty Scale scores 0 through 6. The death rate then rose approximately 5% or more for each point rise in the NH Frailty Scale. The combined death or discharge to a hospital and never return dependent variable followed the same pattern, confirming the expectation that many of those who left to a hospital and did not return were quite clinically complex cases. Here, the rate remained below 20% for NH Frailty Scale scores 0 through 6. Subsequently, the rate rose steadily until, at a score of 14, the rate of those who died or left for an acute hospital and did not return was 65.8%.

The final dependent variable plotted in Figure 2 is the baseline judgment by the MDS assessors as to whether the residents were likely to die in the next 6 months. Here, the rate remained below 6% until a score of 7 on the NH Frailty Scale, at which point the rate began to rise steadily until the projected death rate reached 37% at the upper end of the NH Frailty Scale score.

Figures 3-6 display the rate of dependency across the NH Frailty Scale for residents who were bladder incontinent, had a memory problem, or were dependent in walking, respectively all items that were not included in the independent variable set used to create the NH Frailty Scale. 

Figure 3 specifically displays 3 estimates for bladder incontinence based on the resident’s baseline NH Frailty Scale score as well as on any improvement in that score at follow-up. Two findings stand out. First, the proportion of residents who were incontinent increased for all NH Frailty Scale scores but increased sharply once a score of 4 was attained. Second, an improvement in the NH Frailty Scale score almost always translated into a somewhat lower rate of incontinence. This phenomenon was moderated by the person’s improvement status on the NH Frailty Scale—an improvement of 2 or more points translated into lower incontinence rates from NH Frailty Scale scores 2 through 12. For example, with a score of 6 on the NH Frailty Scale, residents who had experienced no improvement in their frailty score since baseline had a bladder incontinence rate of 65%, whereas residents whose frailty score had improved by 1 point had a bladder incontinence rate of 50%, and those whose score improved by 2 or more points had an incontinence rate of 30%.

Figures 4 and 5 follow the same format as Figure 3 but focus on problem rates for memory recall and walking, respectively. Here, the pattern of findings mimics that of the bladder incontinence rates described in Figure 3. For example, of residents with a walking score of 6 and who did not improve in frailty score, 79% were dependent in walking, whereas for those who had improved by 1 frailty point, the rate was 70%, and for those who had improved by 2 or more points, the rate was 52%.

Figure 6, like Figure 5, looks at walking but uses a somewhat different scoring rule than does Figure 5. Here, it is the proportion of residents who were more dependent at follow-up (ie, whose score declined) than at baseline assessment. While this figure displays the results for decline in walking, the results for decline in bladder incontinence and decline in memory would be similar. As in Figure 5, which looked at dependence in walking at follow-up, the change in walking depicted in Figure 6 confirms a meaningful net benefit for residents who experienced a frailty score improvement. In this case, however, the distinction between those whose frailty score improved by 1 point and those whose score improved by more than 1 point is not as great. For example, of residents with a frailty score of 7 whose score did not improve, 27% declined in walking, whereas for those who had improved by 1 frailty point, the rate was 21%, and for those who had improved by 2 or more points, the rate was 19%.

The NH Frailty Scale was developed from an analysis of more than 100 items from the MDS used in US NHs, with death, death and hospitalization, and clinical judgment of close to death as dependent measures. The resulting 21 scale items represent the domains of function, cognition/communication, medical diagnosis, clinical status, and treatment. Scale scores extended from 0 to 14, with an average frailty score of 4.45 at baseline measurement among 1,097,189 NH residents. The incidence of death within the 120 days following baseline assessment increased steadily for those with a frailty score of 6 or higher. Other researchers have found similar strong associations between higher frailty scores and death.^20-22

Residents who are or who become more dependent in walking, bladder incontinence, and memory at follow-up had increasingly more problematic scores across the categories of the NH Frailty Scale. When we saw an improvement of the NH Frailty Scale score, we also found lower distress rates on each measure, serving to validate the scale. Analyses such as these align with prior work examining gait speed among older adults, including Fried and colleagues’ proposed phenotype of frailty.^8,23-25

The results point to a close association between the frailty scale score and cognitive decline, a finding that is consistent with several previous studies. After completing a 4-year follow-up analysis of 2737 community-dwelling older adults, Auyeung and colleagues²⁶ found an association between physical frailty and cognitive decline. Among 754 noninstitutionalized older adults who were followed over 8 years, Rothman and colleagues²⁷ examined physical ability and weight loss as part of the frailty measure and found that these factors were associated with adverse outcomes. Other works also have demonstrated a close association between frailty and cognitive decline.^28,29

Currently, no universal, population-based frailty screening recommendation exists despite the development of many frailty measures across multiple aging populations.¹⁸ In our view, frailty as an operational construct may be best focused on the experience of population subsets, such as the frailty index specifically for hospitalized older adults evaluated by Evans and colleagues.³⁰ We have created such a frailty scale for NH residents and, earlier, for home care patients.³¹ Others in interRAI have created scales for other populations, such as those in acute hospitals.³²

The NH Frailty Scale meets the recommended goals of being applicable within the context of a comprehensive geriatric assessment and serving as an outcome measure.³³ As a measure of a complex aging concept, the index has potential utility for long-term care planning, medication reviews, and advanced directives including provider and family decisions about life-sustaining treatment. 

Limitations

A secondary analysis design was used to create the NH Frailty Scale, and the data collection was not structured specifically for creating a frailty index. Because we used assessment data gathered from the MDS, we were unable to construct specific assessment items, and we had no control over the choice of measurement scales used for the individual items nor the construction of the chosen scale.

Because of the large number of frail older adults in NHs and other LTCFs, these facilities are best positioned to implement routine frailty screening and assessment and to integrate the results into the care planning process. Use of the NH Frailty Scale, derived from the MDS, represents an efficient approach to monitoring this clinical syndrome among residents in LTCFs. 

Affiliation:
¹ School of Nursing, Bouvé College of Health Sciences,
Northeastern University, Boston, MA
² Institute for Aging Research at Hebrew
SeniorLife, Boston, MA

Disclosures:
The authors report no relevant financial relationships.

Acknowledgements:
The authors remain grateful to interRAI and especially the nursing home residents whose assessments provided the data for this project.

Address correspondence to:
Elizabeth P Howard, PhD, RN, ACNP, ANP-BC, FAAN
School of Nursing, Bouvé College of Health Sciences, Northeastern University
106 H Robinson Hall 360 Huntington Avenue
Boston, MA 02115
Phone: (617) 373-4590
Email: E.Howard@northeastern.edu