The association of geriatric syndromes with hospital outcomes

Hospital Outcomes

The following anonymized variables were identified:

• LOS (days). Prolonged LOS was defined as 10 or more days (no = 0; yes = 1)
• Inpatient mortality (no = 0; yes = 1)
• Delayed discharge (no = 0; yes = 1). This was defined as the total LOS being at least 1 day longer than the LOS up to the last recorded clinically fit date. This date is used in NHS hospitals to indicate that the acute medical episode has finished and discharge-planning arrangements (often via social care providers) can commence.
• Institutionalization (no = 0; yes = 1). This was defined as the discharge destination being a care home, when a care home was not the usual place of residence.
• 30-day readmission (no = 0; yes = 1)

Statistical Analyses

Anonymized data were analyzed with IBM SPSS Statistics (v 22, Armonk, New York) software. Descriptive statistics were given as count (with percentage) or mean (with standard deviation.

To avoid potential problems with multicollinearity in the multivariate regression models, the correlations among the predictor variables were checked using a correlation matrix of 2-sided Spearman’s rho correlation coefficients. Correlations of 0.50 or more were considered large.^15,16

Because all outcomes in the study were binary, multivariate binary logistic regression models were computed. In these models, the odds ratio (OR) reflects the effect size of each predictor; 95% confidence intervals (CI) were requested for each OR. Predictors with P < 0.01 were considered as statistically significant. The classification performance of each logistic regression model was assessed calculating its area under the curve (AUC). 

Sensitivity analyses were conducted after imputing missing data (SPSS multiple imputation procedure) and after fitting interaction terms between geriatric syndromes and discharge by geriatric medicine.

The initial database contained 12,282 nonelective admission and discharge episodes (all specialties) of patients 75 years and older between October 26, 2014 and October 26, 2015. Among those, 8202 (66.8%) were first episodes. Table 2 shows the sample descriptives, and Table 3 shows the breakdown of geriatric syndromes (single and multiple) in the total sample (n = 8282), including missing frailty data.

In the correlation matrix of 2-sided Spearman’s rho correlation coefficients, no correlations with large-effect size were found to suggest issues with multicollinearity; the largest correlation coefficients were between age and CFS (rho = 0.35), HoD and CFS (rho = 0.32), and CCI and CFS (rho = 0.26).

The results of the multivariate regression models are shown in Table 4. The best performing models were the ones for inpatient mortality (AUC = 0.80), followed by institutionalization (AUC = 0.76), and prolonged LOS (AUC = 0.71). After full adjustment, clinical frailty was an independent predictor of prolonged LOS, inpatient mortality, delayed discharge, and institutionalization. HoD was an independent predictor of prolonged LOS, delayed discharge, and institutionalization; and ACS was an independent predictor of prolonged LOS, delayed discharge, institutionalization, and 30-day readmission (Table 4). Results did not significantly change in sensitivity analyses conducted after multiple imputation of missing data and after inclusion of interaction terms (see Supplemental Table 1 and Supplemental Table 2).

Our aim was to study the association of geriatric syndromes (measured in routine clinical care) with hospital outcomes. We found that geriatric syndromes such as clinical frailty, HoD, and ACS were strong independent predictors. Concerning prolonged LOS, delayed discharge, and institutionalization, geriatric syndromes had ORs that were greater than those of traditionally measured factors such as demographics, comorbidity and acute illness severity. Our findings add to the body of knowledge in this area because we accounted for the latter effects. Our experience shows that metrics on geriatric syndromes can be successfully collected in the routine hospital setting and add clear value to the prediction of operational outcomes. This may encourage other hospitals to do the same.

Our findings are consistent with suggestions that accounting for chronic conditions alone may be less informative than also accounting for the co-occurrence of geriatric syndromes.¹⁷ The focus of CFS is on the pre-admission level of physical activity and dependency on activities of daily living, and poorer scores may confer vulnerability to adverse outcomes due to reduced physiological reserve and ability to withstand acute stressors.¹⁸ Other studies have also found CFS to be a good predictor of inpatient outcomes,^19-22 and it has been recommended as a possible means to identify vulnerable older adults in acute-care settings.²³

Our results are consistent with observations that geriatric syndromes are associated with higher risk of institutionalization.²⁹ It was interesting that female gender seemed to be an independent predictor of institutionalization, which is consistent with the results of a systematic review showing that the male-to-female ratio of admission rates ranged between 1 to 1.4 and 1 to 1.6.³⁰

Discharge by general medicine appeared to be associated with a lower likelihood of prolonged LOS, and discharge by geriatric medicine seemed to be associated with a higher likelihood of delayed discharge and institutionalization. Unsurprisingly, geriatric medicine wards tend to absorb the most complex cases, often with complex discharge planning needs.⁸ In that light, CGA in geriatric wards may not be associated with reduced LOS (and it is possible that the LOS of complex patients might have been higher in nongeriatric wards). In addition, inpatient CGA increases frail patients’ likelihood of survival.³¹

Our study suggests that routinely collected metrics on frailty, HoD and ACS may be helpful to better adapt hospital care to the real requirements of aged people. The proportion of older people admitted to acute hospitals with geriatric syndromes continues to increase⁵ and geriatricians are a scarce resource. It will be increasingly important to upskill nongeriatric hospitalists in the recognition and management of geriatric syndromes. Frail older people are becoming the core business of acute hospitals,³² making geriatrics “too important to be left to geriatricians.”³³ Therefore, easily collected metrics on geriatric syndromes may help nongeriatricians identify these syndromes and address them early during admission.

Our study has important limitations. Firstly, geriatric syndromes were not identified with gold-standard measures. For example, ACS in the absence of known dementia should be seen only as a surrogate for delirium. ACS as a proxy measure is likely to underestimate the diagnosis of delirium, because the hypoactive type is commonly missed without valid measures. In addition, a patient with delirium superimposed upon dementia would have been coded as a ‘known dementia.’ The geriatric syndromes’ measures could not be assessed for reliability within the electronic medical records system (eg, regarding sensitivity and specificity against a gold standard, or interrater reliability).

About the potential limitations of CFS, there have been concerns that an interobserver discrepancy in CFS scoring may occur between health professionals. However, 1 study investigated the interrater reliability of CFS between clinicians in 107 community-dwelling older adults 75 years and older, finding a substantial agreement with a weighted k coefficient of 0.76 (95% CI: 0.68 to 0.85).³⁴ Another study reported a CFS-weighted kappa of 0.92.³⁵ Another limitation of CFS in our center is the significant proportion of missing data (28%). As we have shown, missing CFS data are more frequent in situations of very high acuity (including in critical care or surgical areas) or in medical areas when the LOS was short (eg, less than 72 hours).⁸ We tried to address this bias by performing multiple imputation for missing data, which showed similar results.

Another limitation of our study is that we treated geriatric syndromes and the other predictors in the models as independent variables. However, many of the factors may be interrelated, and they present simultaneously in many patients. Indeed, the bivariate correlation between CFS and HoD was of moderate strength, because worsening cognition should score higher on CFS according to the scoring protocol. As expected, there was also a medium-sized correlation between CFS and CCI. It has been suggested that physical and cognitive frailty may be more informative as a single complex phenotype.³⁶ Indeed, the problems of old age tend to come as a package.³⁷

For 30-day readmission, the AUC of the model was small, suggesting the existence of unmeasured explanatory variables. For example, although our results agree that AIS and chronic illness predict readmission,³⁸ the latter still remains an elusive outcome, and a more accurate prediction may be attained by adding socioeconomic variables to models.³⁹Our study echoes the potential utility of incorporating common geriatric clinical features in routine clinical examination and disposition planning for older patients in acute settings.⁴⁰ Hospitals may find it informative to undertake large-scale screening for geriatric syndromes including frailty, dementia, and delirium in all older adults admitted via the ED. When combined with other routinely collected variables such as demographics, AIS, and comorbidity data, this process may provide hospitals with information that will help define the acute needs of the local population and aid in the development of care pathways for the growing population of older adults.