Association of anemia with worsened activities of daily living and health-related quality of life scores derived from the Minimum Data Set in long-term care residents

This study used a retrospective cross-sectional design. Data was obtained from the AnalytiCare LTC database for residents of 27 LTC facilities in the state of Colorado that contributed Minimum Data Set (MDS) version 2.0, laboratory records and pharmacy dispensing records to this integrated database. Although the MDS is a component of the medical chart in LTC facilities, data collected in this instrument may not capture all comorbid conditions noted elsewhere in the chart. The MDS 2.0 has been reported to have moderate to moderate/high validity and reliability, but may underreport some conditions such as depression [28]. Wodchis et al. reported sensitivities exceeding 0.60-0.80 for 22 common diagnoses, but also noted limitations in capturing several conditions [29]. Electronic MDS, laboratory and pharmacy records were complete for all included study residents, though use was not captured of over-the-counter medications not otherwise dispensed by LTC pharmacies. All data used in this study was de-identified, HIPAA compliant, and exempt from requirement for IRB review.

Anemia status was based solely on the Hb value (index Hb) of the lab test performed closest to, and within 90 days, of the index MDS date. Anemia status was defined using the World Health Organization definition [2] (where anemia occurs if Hb <12 g/dL for females and <13 for males). An alternative to anemia status, hemoglobin range, was estimated for each resident by assigning the index Hb to one of five values: ≥13 g/dL, 12 to <13, 11 to <12, 10 to <11, <10 [9]. Estimation of CKD, a covariate in the analysis, was identified from lab or MDS evidence if, from the sCr closest to and within 90 days of the MDS, the resident had either estimated GFR <60 mL/min/1.73 m² (Modification of Diet in Renal Disease equation version 4) [30], or if the resident had renal failure checked on the index MDS Section I1, or else had an entry for chronic kidney disease recorded in MDS Section I3 Other Current of More Detailed Diagnoses and ICD9 codes. Use of certain prescription classes (also used as covariates in the analysis) were based on evaluation of all prescriptions dispensed 30 days prior to and 90 days after the index Hb. All other model covariates and the ADL and HRQOL endpoint measures were derived from the index MDS.

The ADL self-performance ratings in section G of the MDS version 2.0 have been used to evaluate physical functioning in the LTC facility [26, 31]. Carpenter et al. [26] developed a method for estimating ADL performance in the LTC facility by summing the self-performance ratings for seven items in the MDS: bed mobility, transfer, locomotion, dressing, eating, toilet use and personal hygiene. Each item is rated by the LTC coordinator on a scale from 0 (independent) to 4 (total dependence). Alternatively, the coordinator is permitted to assign a value of “activity did not occur during the entire 7 days” to any ADL assessment item. Here, following Carpenter et al. [26] we assigned a substitute rating of 4 (total dependence) by assuming that the resident was incapable of self-performance on this ADL item during the seven-day recall period for this section. We then summed, for each resident, the value of the numeric rating for all seven items, calculating the total MDS-ADL^a performance score ranging from 0 (completely independent) to 28 (completely dependent). Thus a higher MDS-ADL score indicates worse ADL performance.

The HRQOL^b score was calculated by using the MDS-HSI methodology developed by Wodchis et al [27]. These authors developed this method as a practical means of mapping selected items from the standard MDS assessments conducted in LTC facilities and home care settings to an established measure of HRQOL, the Health Utilities Index Mark 2 [HUI2] [32, 33]. The result of this approach yields both an overall HRQOL score, the MDS-HSI, and individual subscores for the six dimensions comprising the HUI2: sensation, mobility, emotion, cognition, self-care, and pain.

Wodchis et al. [27] selected the HUI2 as the basis for the MDS-HSI score, rather than the newer HUI3 instrument, since the HUI2 permits a focus on self-care, considers pain in the context of analgesics to manage it, and places emotion in the context of worry and anxiety rather than happiness vs. depression. The HUI instruments have been administered as both self- and proxy-assessments (e.g. by a spouse or healthcare professional on behalf of the responsible individual) [33]. In their original work, Wodchis et al. [27] found preliminary evidence of convergent and construct validity for the MDS-HSI by first, comparing single-attribute scores from the MDS-HSI with summated scales based on the same attributes, and secondly by separately comparing MDS-HSI scores between residents of supportive housing, recipients of home care services in the community, LTC residents, and patients in a chronic care hospital with HUI2 scores obtained separately from an external reference population (1996 National Population and Health Survey) of community- and institutional-based respondents. A later study [34] assessed criterion validity of the MDS-HSI in a nursing home population against scores derived from interviewer-administered HUI2 in study residents. Wodchis et al. found analogous group-level concordance, but only moderate individual-level agreement. These authors concluded that the MDS-HSI can be used to substitute for the HUI2 in group-level comparisons but not for individual clinical evaluation comparisons [34].

Following the MDS-HSI to HUI2 mapping algorithm reported by Wodchis et al. [27], we calculated, from the index MDS, for each study resident a summary MDS-HSI score and individual subscores for the six MDS-HSI dimensions. The summary score is interpreted such that a value of 1 represents the preference assigned perfect health, while a value of 0 represents dead.

Stata (Intercooled 8.0, College Station TX) was used to conduct statistical analysis. Unadjusted estimates of MDS-ADL and MDS-HSI scores were compared through cross tabulation with both anemia status and hemoglobin range. Multiple regression, adjusted for covariates potentially related to the study endpoints (demographics, renal function, falling history, conditions; a full listing of covariates is shown in Tables 1 and 2), was conducted to separately evaluate the potential independent effect of 1) anemia status, and 2) Hb range, on the study endpoints of MDS-ADL and MDS-HSI scores. Starting with the large set of covariates, backward elimination was used to efficiently reduce the final set of retained covariates, retaining those covariates with P≤0.25 through each successive iteration of the model. Since the WHO definition of anemia incorporates gender with Hb to determine anemia status, the interaction term of gender with anemia status was added to the anemia models.

Performance indicators such as ADL may be relevant predictors of physiologic outcomes in their own right. Abicht-Swensen et al. [37] observed in LTC residents that ADL, as calculated from the MDS, was a strong independent predictor of short-term mortality. Van Dijk et al. [11] found that when ratings for four items in the ADL section of the MDS were added to an adjusted model, that ADL was significantly associated with mortality within the first year of an MDS assessment. This association was independent of anemia and other resident factors.

In the current study the mean MDS-ADL score was 15.5 for the anemic residents and 14.3 for non-anemic residents. This compares with a mean score of 14.8 among mildly cognitively impaired residents and a worsened 19.0 mean score among severely cognitively impaired residents within a single U.S. nursing home reported in Carpenter et al.’s original work on the MDS-ADL [26]. Regarding the potential association between anemia and poorer physical functioning, the evidence for such a link is strong. In the unadjusted analysis in the current study, the mean MDS-ADL summary score was significantly worse for anemic residents. After adjusting for potential confounders in the regression model, we found that anemia was independently associated with greater impairment of MDS-ADL performance. In the adjusted sensitivity analysis substituting Hb range for anemia status, we found that only the moderately severe anemia range of 10 to <11 g/dL had a significantly lower MDS-ADL score when compared with the >13 g/dL reference. The next lowest Hb range of <10 g/dL had a regression coefficient that was similar to the 10 to <11 g/dL range, but the former was not significant since many fewer residents had this severely low value (Table 1).

As described above, Schnelle et al. found in an unadjusted analysis of single-item ADL ratings, that anemic LTC residents with CKD had significantly worse performance than non-anemic CKD residents on every ADL item except for “locomotion off unit,” “eating” and “personal hygiene.” [21] Earlier studies have consistently reported an association between anemia and ADL function for older persons in the community [15‐20].

The mean MDS-HSI score was 0.418 for the anemic cohort (0.411 for females, 0.428 for males) and 0.442 for non-anemic one (0.430 for females, 0.474 for males). Two studies of residents at hospital-based LTC facilities in Ontario have reported substantially lower MDS-HSI scores (0.31 for females, 0.38 for males) [27] and (0.293 for females, 0.286 for males) [25] The somewhat lower MDS-HSI scores found in these earlier LTC studies might be due to high levels of medical impairment noted among these hospital-based cohorts, though MDS-HSI scores from both the current and these earlier LTC studies were both far lower than comparable HUI2 normative scores for the U.S. community population of 0.85 (65–74 years) and 0.83 (75–89 years).

Findings from the current study suggest that there is an association between anemia and impaired quality of life. In the unadjusted analysis in the current study, the mean MDS-HSI summary score was significantly worse for anemic residents. Of the six domains comprising the MDS-HSI summary score, Figure 1 shows that declines in mobility with decreasing Hb level appear to explain much of this difference in quality-of-life assessment between anemic and non-anemic residents. We found, in our adjusted analysis, that anemia was independently associated with a worse MDS-HSI score. In the adjusted sensitivity analysis substituting Hb range for anemia status, we found that only the severe anemia range of <10 g/dL had a significantly lower MDS-HSI score when compared with the >13 g/dL reference.

Several earlier studies have evaluated the association between quality-of-life measures and anemia. As described above, Lam et al. found that anemia had a significant association with HRQOL in the LTC setting [25]. That study assessed HRQOL with the same MDS-HSI measure used in the current study, though in a different LTC population (hospital-based LTC facility in Ontario) and utilizing a different regression modeling approach (full model retaining 60 disease predictors). In that study anemia was significantly associated with only 0.022 unadjusted and 0.006 lower adjusted MDS-HSI score vs. the significantly lower scores of 0.024 (unadjusted) and 0.034 (adjusted) that we found in the current study of a pool of U.S. nursing homes and utilizing a stepwise regression model. Studies outside of the LTC setting have also reported an association of anemia with HRQOL in older patients [22‐24].

In the current study, the independent effect of anemia was a 1.62 point worse adjusted MDS-ADL score. Findings from this study also showed that anemia was associated with a 0.034 adjusted lower MDS-HSI summary score. When modeling relationships between a condition such as anemia and worsened performance on humanistic measures at a single point in time, one concern is interpretation of the meaning of such differences. A limitation of the current study is that one can only speculate regarding such meaning; meaningful differences are not formally assessed in such between-group models [38‐40]. Minimally-important individual change differences have been assessed, but the latter are not directly comparable to between-group model effects. For instance, Carpenter et al. evaluated changes in the MDS-ADL within individuals and concluded that ”[a] change of one point in the MDS-ADL scale denotes a clinically meaningful change.” A difference of 0.03 for the Health Utilities Index (HUI) instruments has been identified by Drummond [41] and Grootendorst et al. [42] as a clinically-important change. Drummond [41] has stated that a difference as low as 0.01 may be may be meaningful and important in some settings.

This study has several additional limitations. First, even though the LTC facility coordinator conducting the MDS assessment is directed to observe and communicate with the resident when completing the assessment, data is gathered from multiple sources [43], thus assessments of MDS-ADL and MDS-HSI are largely by proxy, rather than by self-report, and so are dependent on the quality of these data sources. Second, although the MDS prompts the LTC facility coordinator to report, in detail, diseases and conditions currently experienced by the resident (either through checkbox or open-ended entry of conditions), underreporting of such diseases and conditions on the MDS has been known to occur [44]. However, other studies have found high inter-rater reliability of quality indicators and strong correlations with independent observation of residents [45, 46]. Third, potential confounders with anemia in the adjusted regression models we were limited to only those data elements contained in the MDS, pharmacy fills or index lab values. Other relevant factors potentially related to the clinical endpoints studied may not have been included in this study. Fourth, the regression models only tested a linear association between anemia/hemoglobin range and the outcomes of MDS-ADL and MDS-HSI score. Non-linear regression models (e.g. quantile regression) or transformations of these scores might have revealed different findings. Finally, anemia status was determined by a single index Hb value, so we were unable to determine whether the anemia status was an acute or chronic state, what the cause of the anemia was, and whether the MDS-ADL and MDS-HSI assessments changed over time within residents as anemia status and Hb levels changed.

^aCarpenter et al. did not assign a name to their summary ADL measure. We introduce the term “MDS-ADL” here.

^bAlthough some researchers regard health status and HRQOL as separate constructs, these terms are used interchangeably in the original work of Torrance et al. [Torrance GW, Feeny DH, Furlong WJ, Barr RD, Zhang Y, Wang Q. Multiattribute utility function for a comprehensive health status classification system. Health Utilities Index Mark 2. Med Care. Jul 1996;34(7):702–722].