The impact of early comprehensive geriatric screening on the readmission rate in an acute geriatric ward: a quasi-experimental study

This study was conducted in the geriatric ward of the National Taiwan University Hospital. The geriatric ward serves elderly patients with acute medical illness requiring admission.

This quasi-experimental prospective study included a historical comparison group and a post-intervention group.

A total of 415 patients admitted from July 2011 to June 2012 were included in the historical comparison group. They were randomized selected from the 848 patients admitted to geriatric ward in a one-year period. The CGA was performed on selected frail inpatients (in 40 of 415 patients, about 9.6% of controls). We had reviewed the medical chart and confirmed the patients discharged alive. The same patient would be enrolled once, even if he or she had admission several times in this year. We also check the patients’ medical chart at outpatient department (OPD) within 2 months after discharge. If the patient loss follow up at OPD in this period would label as censored case.

All patients were elderly (age > =65) at admission and under routine geriatric medical care, including functional evaluations, disease treatments and care-skill educations.

Participants were excluded if they died prior to discharge or were transferred to another medical department. In the CGS group, written informed consent was obtained from participants or from their legal guardians if the patients had serious cognitive impairment. The study was approved by the Institutional Review Board of the National Taiwan University Hospital (No: 201303058RINC).

Data collected from medical charts included demographic data, disease comorbidities, functional status and health utilization. Participants in the CGS group received the CGS within the first 2 days of admission. The CGS used simplified but valid screening tools, e.g., the Confusion Assessment Method (CAM) criteria for delirium [19], the core symptoms for depression [20], and the Mini-Cog for dementia [21]. It also maintained the core metrics of the CGA, which consisted of measures of delirium, depression, dementia, visual and hearing capacity, physical performance, falls, polypharmacy, pain, pressure sores, incontinence, tubes, nutrition, caregiver issues, and socioeconomic issues. The CGS was administered by the same study nurse who received geriatric skill training before the study, and she accompanied our experienced nurse practitioners during the first 2 weeks of practice. The entire evaluation took approximately 10–15 min depending on the patient’s condition. All the results were provided to the attending physician of the geriatric ward. If the patient had positive CGS results, a complete CGA was conducted for further evaluation and management (Additional file 1: Table S1). Patients received non-pharmacologic or pharmacologic management, rehabilitation, nutrition modification, fall prevention and care-skill education according to their needs and the physician’s suggestion.

All patients were followed for 30 days after hospital discharge by medical chart review and/or telephone calls. The main outcome was unexpected patient readmission within 30 days after discharge. Planned readmissions, such as admissions for cancer chemotherapy, scheduled operations or scheduled rehabilitation, were defined as scheduled admissions and were not included as part of the main outcome. Patients who died within 30 days at home or in an emergency room after discharge were excluded from the analysis.

According to previous admission data, the required sample size was calculated before the study to obtain 80% power at the 5% significance level [22].

T-tests (for normally distributed continuous variables) and chi-square tests (for categorical variables) were used for intergroup comparisons.

Curves for the probability of no readmission within 30 days were created with the Kaplan–Meier method and compared using the log-rank test. A multivariate analysis for possible significant factors was performed using the Cox proportional regression model. We developed several models to test the stabilization of the effect of CGS.

Stratified analyses were performed using age, gender, the Charlson comorbidity index (CCI), the Barthel index score, and number of admissions within 1 year to assess the association between the CGS and the 30-day readmission risk.

Finally, we used a scoring system to identify the patients who received the most benefit from the CGS. The Kaplan–Meier method and the log-rank test were used to analyse the differences between the scores.

After excluding patients who died during the study course or were transferred to other medical departments, a total of 380 controls and 377 patients who received the CGS were included in our final cohort study (Fig. 1). There were 9 patients with scheduled admissions (3 patients in the control group and 6 patients in the CGS group) within 30 days of discharge. The number of scheduled admissions in the two groups was not significantly different (p = 0.50), and all of them were censored at their scheduled admission date. Three of the patients in the CGS group died at home or in an emergency room before readmission within 30 days of discharge and were excluded from the final analysis.

The CGS group had fewer patients with a history of hypertension and dementia but more patients with a history of stroke than the historical group. Age, gender, other comorbidities, and functional status were similar between the two groups. The 30-day readmission rate was significantly lower in the CGS group than in the historical group (11.4% vs 16.9%, p = 0.03, Table 1).

We also examined the characteristics of the patients who were readmitted within 30 days and compared them with those who were not (Additional file 2: Table S2). Patients who were readmitted tended to have a longer lengths of stay, more comorbidities, poorer functional status at admission (Barthel index score ≤ 35), and more prior admissions within the previous year than patients who were not readmitted. Regarding specific diseases, patients with congestive heart failure, atrial fibrillation, history of peptic ulcer, chronic kidney disease, and metastatic cancer were at a higher risk of 30-day readmission.

The probability of being readmission-free within 30 days of discharge between the CGS and historical groups was analysed using the Kaplan–Meier method (Fig. 2a), and the p-value by the log-rank test was 0.03. Before the proportional hazards analysis, we tested the time interaction model between the time variable and study groups with no significant finding (p = 0.60). Different models were developed to confirm the effect of the CGS (Table 2). We adjusted age and gender in model 1, and we added the disease factors (Charlson comorbidity index), functional status (Barthel index score) and medical utilization (length of stay and number of admissions) separately to the analysis. Model 5 was a completed, adjusted model for the above factors. The hazard ratio of the CGS for the different models ranged from 0.64 (95% CI 0.43–0.94 in model 4 and 0.43–0.95 in model 5) to 0.67 (95% CI: 0.45–0.98 in model 1 and 0.45–0.98 in model 3). In addition, several factors had significant effects on our readmission model. A higher Charlson comorbidity index resulted in a higher readmission risk in model 2 (HR: 1.85 95% CI: 1.24–2.75) and model 5 (HR: 1.64 95% CI: 1.09–2.46). Additionally, patients with a lower Barthel index score at admission, greater length of hospital stay or prior admission history within 1 year had a higher risk of readmission (Table 2).

Stratification with key variables was important to identify the benefit of an early CGS for readmission. Age, gender, CCI, Barthel index score, and number of admissions within the prior year were chosen for stratification. The model was fully adjusted except for the stratification variables, and the results are shown in Table 3. The CGS had a significant benefit for patients <=80 years of age (HR: 0.35), women (HR: 0.51), those with a Barthel index score < =35 (HR: 0.53), and no admission history within the prior year (HR: 0.52). In the crude and basic models, the CGS showed benefits for patients with high comorbidity (CCI > =3), but this was not the case in the fully adjusted model.

Based on the stratification variables, we scored the potential benefit to patients using the following factors: age < =80, female gender, CCI > =3, BI<=35, and no admissions within the prior year. We divided the patients into 2 groups, a low-potential group with individual scores of 0–2 and a high-potential group with scores of 3–5. The results of the two groups were analysed using the Kaplan–Meier method (Fig. 2b), and the log-rank test showed that the p-value was 0.98 in the low-potential group and 0.001 in the high-potential group.