Effects of Comprehensive Geriatric Care Models on Postoperative Outcomes in Geriatric Surgical Patients

A Systematic Review and Meta-analysis

Aparna Saripella; Sara Wasef; Mahesh Nagappa; Sheila Riazi; Marina Englesakis; Jean Wong; Frances Chung


BMC Anesthesiol. 2021;21(127) 

In This Article


Search Results

A complete search of the selected articles is summarized in Figure 1, in accordance with the PRISMA statement.[25] A total of 35,186 articles was identified. After applying the deduplication process, 8843 articles were removed. The titles and abstracts of the remaining 26,343 articles were screened for selection criteria, after which 20 articles remained. Full-text screening of these 20 articles resulted in 11 articles which were included for the qualitative synthesis of the review.[10,11,13,15–19,22–24] We excluded nine articles due to no care pathway, no surgery, a non-geriatric population, or incorrect types of article.

Figure 1.

PRISMA study flow diagram

Patient and Study Characteristics

Eleven studies (RCTs: 4; non-RCTs: 7) included with 2672 patients (intervention group n = 1383, control group: n = 1289).[10,11,13,15–19,22–24] An overview of the study and patient characteristics are summarized in Table 1. Most of the included studies are from Europe [Netherland (n = 3), United Kingdom (n = 1), Spain (n = 1), Sweden (n = 1),[13,17–19,23,24] and United States (n = 4),[11,15,16,31] while one is from China (n = 1).[10] Patients in the included studies underwent cancer (n = 4),[13,18,19,23] abdominal (n = 3),[10,15,16] vascular (n = 2),[11,17] spinal (n = 1),[22] and total hip arthroplasty (n = 1)[24] surgeries. The included articles had the following study designs: RCT (n = 4),[10,13,17,23] prospective cohort (n = 2),[11,15] retrospective cohort (n = 3),[16,19,22] as well as pre-intervention and post-intervention study design (n = 2).[18,24]

Study Quality Assessment – Risk of Bias

Using the Cochrane tool, the four RCT's had low bias on most of the domains. Random sequence generation and incomplete outcome data were the most sufficiently addressed, with all four studies reporting low bias in these domains. Allocation concealment and blinding of outcome assessment were the least sufficiently addressed. In allocation concealment domain, two studies reported low bias and two reported high bias. Similarly, two studies reported low bias and two reported unclear bias in outcome assessment domain.[32] (S-Table 1) According to the Newcastle Ottawa scale scoring system, the quality of the six non-RCTs ranked from 7 to 9 indicated low risk of bias.[30] One study was considered being of high risk due to selection and outcome bias.[18] (S-Table 2 and Table 3).

GRADE Evaluation

GRADE evaluation of the quality of evidence for the outcomes: prevalence of delirium, LOS, 30-days readmission, and 30-days mortality was conducted. The quality of evidence of the RCTs and non-RCTs together on delirium, LOS, 30-days readmission, and 30-days mortality was rated very low due to risk of bias and imprecision, respectively. (S-Table 4).

The Different Comprehensive Geriatric Care Models

Overview of the different comprehensive geriatric care models is summarized in Table 2. All geriatric care models contained CGA, which is an established multi-domain assessment addressing patients' physiological, social, psychological, and functional state before surgery.[17] The primary feature is cognitive status screening and intervention protocol directed to cognitive impairment. Proactive care of Older People (POPS)[17] model referred patients to specialists following diagnosis of cognitive impairment or delirium, while the majority followed recommendations from caring physicians based on the initial cognitive assessment.[11,13,15,19,22–24] Some studies did not have cognitive impairment intervention,[16,24] while others have devised models for addressing impairment, such as orientation communication in the Hospital Elder Life Program (HELP).[10,14] Intervention tools to specifically address frailty/functional status were included in all the models, with one exception.[16] Intervention tools either took the form of exercise regimens,[10,14,19] or tailored plan determined after assessment,[13,15,17,22,23] while one study referred to general rehabilitation efforts.[18]

Post-operative Outcomes

Delirium. Six studies consisting of 916 patients in the intervention group and 695 patients in the control group reported the prevalence of delirium. Even though the prevalence of delirium was 2.1% less in the intervention group compared to the control group, it was not significant statistically (13.8% vs 15.9%; OR: 0.76; 95% CI: 0.30–1.96; I2: 89%; p = 0.57) (Figure 2a). Our influential analysis showed that McDonald et al. 2018 contributed the maximum heterogeneity. When this study was removed and pooled prevalence of delirium was recalculated, the heterogeneity decreased by 75%, and the prevalence of delirium was significantly less in the intervention group compared to control group (10.2% vs. 18.6%; OR: 0.44; 95% CI: 0.30–0.64; I2: 14%; p < 0.0001). We conducted the subgroup analysis based on the type of study (RCTs vs. non-RCTs). The prevalence of delirium was reported in three RCTs,[10,13,17] and three non-RCTs.[15,19,22] Among the RCTs,[10,13,17] the prevalence of delirium was significantly less in the intervention (n = 430) compared to the control group (n = 420) (7.9% vs. 16%; OR: 0.45; 95% CI: 0.29, 0.70; I2: 0%; p = 0.0003). The absolute risk reduction for the prevalence of delirium is 8.28% (95% CI: 3.9, 12.6), and the number needed to treat is 13 (95% CI: 7.9, 25). The pooled estimate remained significant, and heterogeneity remained at zero after conducting the sensitivity analysis for this significant outcome. Among the non-RCTs,[15,19,22] there was no significant difference between the two groups on the prevalence of delirium (intervention (n = 486) vs. control: (n = 275) (19% vs. 16%; OR: 1.33; 95% CI: 0.17, 10.56; I2: 95%; p = 0.79).

Figure 2.

Forest plot displaying a meta-analysis of the delirium and LOS

Hospital Length of Stay (LOS). Eight studies reported data on LOS, three from RCTs,[10,13,17] and five from non-RCTs.[15,16,19,22,24] However, two RCTs were excluded from meta-analysis due to clinical heterogeneity as a lot of patients stayed in ICU postoperatively.[10,13]

In the pooled estimate from six studies (five non-RCTs and one RCT) with different care models, there was no significant difference in the LOS between the intervention (n = 799) and control groups (n = 646) (mean difference: -0.55; 95% CI: − 2.28, 1.18; I2: 93%; p = 0.53)[15–17,19,22,24] (Figure 2b).

30-days Re-admissions. Seven studies consisting of 884 patients in the intervention group and 704 patients in the control group reported on 30-days re-admission. Overall, there was no significant difference in the 30-day readmission rates. The 30-day re-admission rates were 12.1% in the intervention group compared to 14.3% in the control group (OR: 1.09; 95% CI: 0.67–1.77; I2: 50%; p = 0.73) (Figure 3a). Subgroup exploration based on the type of study (RCTs vs. non-RCTs) did not show any significant difference in the 30-day readmission rates. Out of seven studies, two RCTs measured the rate of the 30-day readmission rates (intervention vs. control: 18% vs.14%; OR: 1.35; 95% CI: 0.81, 2.25; I2: 0%; p = 0.25)[17,23] and five non-RCT studies provided data on the 30-day readmission rates (intervention vs. control: 10% vs. 14%; OR: 0.98; 95% CI: 0.48, 2.03; I2: 60%; p = 0.96).[15,16,18,19,22]

Figure 3.

Forest plot displaying a meta-analysis of 30-days readmission and 30-mortality. Abbreviations: RCT, Randomized Controlled Trials; Non-RCT, Non- Randomized Controlled Trials; LOS, Length of hospital stay

30-days Mortality. Five studies consisting of 720 patients in the intervention group and 604 patients in the control groups reported on 30-day mortality. The pooled data on 30-day mortality was not significantly different between the intervention vs. control group (3.2% vs. 2.1%; OR: 1.34; 95% CI: 0.66–2.69; I2: 0%; p = 0.42). Out of five studies, only one RCT reported data on 30-day mortality (6.8% vs. 2.6%; OR: 2.63; 95% CI: 0.80, 8.57; p = 0.11)[13] and the pooled data from the four non-RCTs did not show significantly difference between the intervention vs. control group (2% vs. 2%; OR: 0.93; 95% CI: 0.39, 2.21; p = 0.86)[15,16,18,19] (Figure 3b).

Other Postoperative Outcomes. S-Table 5 contains detailed secondary outcomes. Complications such as pneumonia (n = 3),[15,17,22] discharge to home with self-care (n = 2),[15,17] activities of daily living (ADL) (n = 2),[23] functional status at 30 days,[11] and geriatric syndromes, and events[19] were reported in some studies. Three studies reported no significant difference in pneumonia.[15,17,22] The percentage of patients discharged to home with self-care was reported in two studies,[15,17] with one study reporting a significant difference of p = 0.04.[15] The study reporting ability to achieve ADL did not show any improvement, but functional status at 30 days was significantly improved (p < 0.01). Similarly, geriatric syndromes and events improved significantly with a p-value < 0.001.[19]