Critical Appraisal of Marsden et al. (2022) – 

“Translation of the Geriatric Emergency Department Intervention into other Emergency Departments: A Post Implementation Evaluation of Outcomes for Older Adults”

Article Summary (Title & Purpose)

Marsden et al. (2022) conducted a post-implementation cohort study to evaluate a geriatric-focused emergency care model known as the Geriatric Emergency Department Intervention (GEDI). The GEDI is a nurse-led, physician-championed ED intervention designed to improve outcomes for frail older patients in the emergency department . This study aimed to assess the effectiveness of implementing the GEDI model in two Australian hospitals (one large teaching hospital in North Queensland and one medium teaching hospital near Brisbane) for patients aged 70 and above . In particular, the authors compared key outcomes before vs. after introducing GEDI, and also between patients who received the GEDI team’s care vs. those who did not during the post-implementation period . The outcomes examined included disposition (admission vs discharge), emergency department (ED) length of stay, hospital length of stay, 28-day in-hospital mortality, 28-day ED re-presentation, 28-day hospital readmission, and healthcare costs . Overall, the purpose was to determine if the translation of this geriatric ED model into new settings would lead to improved outcomes and cost benefits for older adults.

CASP Cohort Study Appraisal

1. Clear Focus

Focused issue/question: The study had a very clear focus. It addresses a well-defined problem – improving emergency care outcomes for older adults – and evaluates a specific intervention (the GEDI model) intended to tackle this issue. The population (ED patients ≥70 years old), the intervention/exposure (implementation of the GEDI model, a specialized geriatric ED team), and the outcomes of interest were explicitly stated. The authors’ overall aim was “to evaluate the effectiveness of the implementation of the GEDI model of care into two EDs” , focusing on measurable outcomes like admissions, lengths of stay, re-presentations, mortality, and costs. This clear statement of purpose and predefined outcomes indicates a well-focused research question. The importance of the issue is also established: with more frail older patients presenting to EDs, models like GEDI are innovated to reduce poor outcomes . In summary, the study’s objective was clearly formulated and relevant to geriatric emergency care.

2. Recruitment of Participants

Recruitment method: Instead of recruiting individual patients, this cohort was defined by time and setting – essentially all eligible older patients presenting to the ED in specified periods. This method is appropriate and minimizes selection bias, since every patient aged 70 or above who visited the two study EDs during the study windows was included . Specifically, de-identified records for all patients ≥70 at Hospital A (Jan 2017–Dec 2018) and Hospital B (July 2016–Sep 2018) were extracted from hospital databases . By including entire populations of older ED patients pre- and post-implementation, the study captured a representative sample of the target population. There was no indication of exclusions (aside from omitting a brief transition period at Hospital B when GEDI staff were only partially in place , to ensure a clean comparison). This comprehensive inclusion makes the results more generalizable. The hospitals were selected based on willingness and funding to adopt GEDI (a form of purposeful site selection), but within each site the coverage of patients was complete. Overall, participant inclusion was appropriate and likely free of selection bias, as it encompassed all eligible patients in those EDs and timeframes.

3. Measurement of Exposure (Intervention)

Exposure assessment: The “exposure” in this cohort study is the GEDI model intervention, assessed in two ways: (a) being in the post-GEDI implementation period (the ED environment where the model was active) versus the pre-implementation period, and (b) at an individual level in the post period, whether a patient was seen by the GEDI team or not . These exposures were clearly defined and measured using hospital records and implementation timelines. The study clearly delineated the pre- vs post-GEDI periods for each hospital and even excluded a partial-implementation phase to avoid misclassification . In the post period, the hospital records indicated if a patient received care from a GEDI nurse/clinician, allowing classification of patients as “seen by GEDI” or “not seen by GEDI” . This method of exposure measurement is objective and reliable – it likely relied on documentation by the GEDI team or flags in the ED system. Misclassification is therefore minimal (one would know if the specialized geriatric team saw the patient). It’s worth noting that GEDI nurses worked only certain hours (daytime/evening shifts) , so patients arriving overnight, for example, would systematically fall into the “not seen by GEDI” group. The study did capture time of presentation in the data and acknowledged this operational detail, but it did not explicitly adjust for time of day in analysis. Nonetheless, by defining exposure based on actual implementation status and GEDI contact, the authors ensured a valid measurement of the intervention. In summary, the exposure (implementation of GEDI) was well-defined and accurately ascertained from routine data, reducing information bias. One remaining consideration is that assignment to “seen by GEDI” was not random – GEDI staff prioritized the frailest patients – which introduces potential confounding by indication (addressed below).

4. Measurement of Outcomes

Outcome assessment: The study evaluated multiple clinically relevant outcomes, all of which were measured using hospital administrative and clinical data systems. These outcomes included:

• Disposition: whether the patient was admitted to hospital or discharged home from the ED .
• ED length of stay (LoS): time in hours/minutes spent in the emergency department .
• Hospital LoS: length of inpatient stay (for those admitted), in days .
• In-hospital mortality within 28 days: any all-cause death occurring in-hospital within 28 days of the ED visit .
• ED re-presentation within 28 days: return visits to the ED (for any reason) within 28 days after discharge .
• Hospital readmission within 28 days: unplanned admissions within 28 days after the initial discharge (for those originally discharged) .
• Healthcare costs: including ED costs per presentation and in-hospital costs per admission, derived from the hospitals’ financial systems (using DRG-based costing) .

All these outcome measures were obtained from electronic records, ensuring objectivity (e.g. timestamps for lengths of stay, discharge disposition recorded in ED systems, mortality from hospital records, cost from finance data). Because the data were collected as part of routine hospital operations, measurement error is likely low. There was consistency in how outcomes were defined pre- and post-implementation (e.g. the way admissions or re-presentations were recorded did not change). The follow-up period for repeat visits was limited to 28 days, which was clearly stated and appropriate to capture short-term outcomes of the ED intervention . One minor limitation is that out-of-hospital outcomes (like deaths outside the hospital or presentations to other hospitals) would not be captured, but the study focused on in-hospital and same-site outcomes by design. Overall, the outcome measurements can be considered reliable and valid, and were applied equally to both comparison groups, minimizing measurement bias.

5. Confounding Factors

Confounder identification and control: The authors recognized that, as an observational cohort study (with a before-after design and non-random allocation to the GEDI service), confounding could influence results. They took steps to identify important confounding variables and adjust for them in the analysis. In their multivariable regression models, they controlled for a range of factors “based on expert knowledge of the system” , including: age of the patient, gender, triage acuity (Australasian Triage Scale category), mode of arrival (ambulance or other), and major diagnostic categories (specifically flags for trauma-related presentations and cardiac presentations) . These factors were appropriate since they could affect outcomes independently of the GEDI (for example, sicker patients with higher triage acuity or arriving by ambulance are more likely to be admitted regardless of GEDI, trauma cases might have different dispositions, etc.). By adjusting for these, the study aimed to isolate the effect of the GEDI model.

It appears the researchers did not adjust for certain other potential confounders explicitly, such as baseline comorbidity burden or a direct measure of frailty. The GEDI team did use a frailty screening tool (interRAI ED Screener) to prioritize patients , which implies that patients “seen by GEDI” were likely more frail or complex than those not seen. This kind of confounding by indication is hard to fully remove; however, the authors indirectly addressed it by including age and some diagnostic indicators as proxies, and by analyzing outcomes both in the overall population and within the post-implementation group. Another potential confounder is time-related trends (for example, any general improvements in hospital processes or policies like the National Emergency Access Target during the study period). The study did not have a separate control hospital without GEDI, so any secular trend could bias the pre/post comparison. The authors acknowledge, for instance, that Hospital B experienced an 11% increase in population ≥70 in its catchment during the study, which led to higher ED demand and possibly shorter hospital stays due to bed pressure . Such context suggests external factors may have influenced outcomes alongside GEDI.

Despite these considerations, the multivariate modeling approach and the inclusion of two different hospitals strengthen confidence that major confounders were accounted for. In summary, key confounding factors were identified and adjusted, though some residual confounding is possible. The authors’ discussion reflects awareness of these issues, lending transparency. Overall, the strategies to manage confounding were solid for a cohort study design, but we should remain cautious in attributing causation given the non-randomized nature.

6. Completeness of Follow-Up

Follow-up and attrition: This study leveraged existing records, so traditional “loss to follow-up” does not apply as it would in a prospective cohort. Essentially, all included patients had their outcomes tracked through the hospital information systems. The follow-up for outcomes like re-presentation and readmission was 28 days, which was uniform for all groups . If a patient did not return within 28 days, they were considered as having no re-presentation (censored at 28 days in survival analysis) – this approach was correctly handled by survival methods. Because the data were obtained from hospital databases, outcomes such as whether the patient was admitted, how long they stayed, and in-hospital death are fully captured for every case. There was no indication of missing outcome data in the results; for example, the number of ED visits analyzed pre and post (Table 1) matches the numbers of patient records retrieved , implying no attrition from the dataset. The linkage between ED records and inpatient records was done by data managers, which would capture outcomes like 28-day readmissions within the same health service .

A possible limitation is that if patients sought care at a different hospital after their index visit, such events wouldn’t be counted as “re-presentations” or “readmissions” in this study. However, this should not create bias between the study groups, as there’s no reason to suspect one group (pre vs post, or seen vs not seen) would systematically go to other hospitals more often. All in all, follow-up was essentially complete for the defined outcomes. The use of routine data avoided drop-outs, and the 28-day outcome window was the same for everyone. We can be confident that outcome data completeness was high, and the results were not diluted by missing follow-up information.

7. Results – Clarity and Credibility

Clarity of results: The findings are presented clearly and in an organized manner. The authors split results by hospital (Hospital A and B) to account for any site-specific differences, which enhances clarity. They used appropriate statistics for time-to-event outcomes (survival analyses with cumulative probability curves for discharge vs admission) and regression models for other outcomes, presenting the results with hazard ratios, adjusted mean differences, and 95% confidence intervals. Key results are summarized in tables and figures, making it easy to see the differences associated with the GEDI intervention. For example, they provide a figure showing the cumulative probability of admission or discharge within 24 hours, which visually demonstrates quicker discharges in the post-GEDI period . The text explicitly states the direction and magnitude of effects. In Hospital A, there was an increased hazard (faster likelihood) of discharge both in the post-GEDI period and for patients seen by a GEDI nurse , and correspondingly fewer admissions; meanwhile time to admission (for those who ended up admitted) was slightly longer in the post period . They report that older age slowed discharge (as expected), and that the post-GEDI period was associated with a reduced chance of 28-day ED re-presentation (HR ~0.88) . They then detail secondary outcomes like hospital length of stay and mortality clearly. The results are thus communicated in a logical sequence, with enough detail for the reader to verify the authors’ interpretations.

Credibility of results: The results appear credible and plausible. The trends observed (more discharges home, shorter stays in ED, etc.) are consistent with what one would expect if a geriatric-focused intervention improved care coordination and decision-making. Importantly, the study’s large sample size (nearly 19,000 older patient visits at each hospital) lends weight to the statistical credibility of the findings . The use of an independent statistician to perform the analysis also adds to credibility by reducing potential analytic bias from the research team. The consistency of positive outcomes across two different hospitals further supports that the effects are real and not just a fluke or site-specific anomaly. Additionally, the authors have been cautious in interpreting results – for instance, noting that while the risk of in-hospital death decreased in the post-GEDI group, the absolute number of deaths was small , and offering explanations for unexpected findings (like why patients seen by GEDI had slightly longer hospital stays yet lower costs ). Their discussion aligns the results with known evidence and rationales. No glaring inconsistencies or implausible findings are present; even the reduction in mortality, which is a strong outcome, is believable in context (the GEDI model may prevent unnecessary admissions or complications, thereby lowering in-hospital deaths). In summary, the results are well-presented and believable. The clear reporting with effect sizes and CIs allows readers to judge the significance, and the findings make clinical sense, which together enhance confidence in the credibility of the results.

8. Precision and Statistical Significance

Precision of estimates: Thanks to the large number of patients, the study’s estimates are quite precise. Most outcomes are reported with 95% confidence intervals that are reasonably narrow, indicating a high level of statistical power. For instance, in Hospital A the hazard ratio for time to discharge in the post-GEDI period was around 1.13 (implying faster discharges), and for patients seen by GEDI it was ~1.30; these estimates had tight CIs that did not cross 1.0 . Likewise, the hazard ratio for 28-day re-presentation in the post period was 0.88 with CI 0.85–0.92 , a narrow interval showing a statistically significant reduction in return visits. An example from the text: “there was a significant increase in the time to admission during the post-GEDI period (Hazard ratio 0.88, CI 0.85–0.92) and if seen by a GEDI (Hazard ratio 0.76, CI 0.72–0.81)” – these precise estimates (note the small margin between the CI bounds) reflect high confidence that the true effect is not zero. The authors often present results as differences with confidence intervals; for instance, they found the cost per hospital admission decreased by $88 on average in the post-GEDI period compared to pre (95% CI from –$270 to $99, which narrowly includes zero indicating a possible modest effect) . Where the CI did cross the null (as in that particular cost example), the authors are transparent about the uncertainty (effectively noting it might not be a significant change). Many other outcomes (like mortality risk, ED length of stay) had clearly significant differences with tight CIs, given the large sample.

Believability of statistics: The analysis methods used (survival analysis for time-dependent outcomes, regression for others) are appropriate and lend validity to the significance reported. P-values aren’t explicitly listed in the text we have, but the confidence intervals tell the story of significance. There is no evidence of “p-hacking” or selective reporting; the paper reports on all stated outcomes, whether significantly changed or not (for example, 28-day readmission rates were mentioned even if they might not have changed much, and cost outcomes are reported with nuance of significance). The consistency in direction of effect (all pointing towards better outcomes or no harm with GEDI) also adds believability that the model had a beneficial impact. In sum, the results are statistically robust and the inferences drawn (e.g., that GEDI likely contributed to improved outcomes) are justified by the data. The precision of the estimates strengthens our confidence that these findings are not due to chance variation.

9. Applicability to Local Population/Context

External validity (generalizability): The study was conducted in Queensland, Australia, and its findings are highly pertinent to similar healthcare settings, especially in Australia and New Zealand. Both study sites were public teaching hospitals – one large urban hospital and one medium regional hospital – which are comparable to many hospital settings in Australasia. The patient population (aged 70+ presenting to ED) is one that all Australian/New Zealand hospitals manage, and their characteristics (average age ~79–80, mix of medical and trauma cases) would be common in other hospitals as well . Because the GEDI model was implemented in real-world conditions within the existing hospital staffing and processes (with some local adaptations), the outcomes likely reflect what could be achieved in other EDs that adopt a similar model. The results showed improvements even when the model was translated to new sites, suggesting it’s not an effect unique to one hospital’s context . This boosts confidence that other hospitals in Australia or NZ could see comparable benefits for older patients if they implement a geriatric ED intervention.

A consideration for local applicability is that both hospitals had supportive conditions: they had leadership buy-in and external funding from the state health department to implement GEDI . In a local context where resources or buy-in are lacking, the same outcomes might not materialize to the same extent. However, the study demonstrates that even with some challenges (Hospital A lost its ED physician champion early on, for example, yet continued the program with adaptations ), the model still conferred benefits. Australia and New Zealand share similarities in emergency care structure and face the same demographic pressures of an aging population, so it’s reasonable to apply these findings locally. Clinicians and administrators in local EDs can take these results as evidence that focusing on geriatric-specific care processes can improve throughput and outcomes for older patients. In summary, the study’s context and population are sufficiently similar to the local setting, making the results applicable to geriatric emergency care in Australia/NZ. Any minor differences in local context (such as funding models or workforce) would need consideration, but the core concept of GEDI has demonstrated transferable value.

10. Overall Value of the Findings

Value and implications: This study provides valuable evidence that implementing a dedicated geriatric care model in the ED can lead to meaningful improvements in patient outcomes and system performance. The findings are practically significant: the GEDI model was associated with more patients being safely discharged home (avoiding potentially unnecessary admissions), shorter stays in the ED (improving patient flow and comfort), reduced in-hospital mortality for older adults, and indications of cost savings for inpatient care . These outcomes address critical goals in emergency medicine and geriatric care – namely, to reduce harms that can come from prolonged ED stays or hospitalizations for older people (such as delirium, hospital-acquired complications) and to use healthcare resources efficiently. The fact that the model achieved these benefits when rolled out in new hospitals (not just in a single trial site) is especially valuable. It suggests that the GEDI model is robust and adaptable, providing a template for other hospitals to follow. The study also adds to the field of implementation science by demonstrating that with the right support and adaptation, an intervention proven in one setting can be translated to others with positive outcomes .

From a healthcare system perspective in Australia/New Zealand, these results are timely and relevant. They give decision-makers evidence that investing in geriatric ED teams can pay off in both patient and financial terms. The authors note that improvements at the new sites were somewhat “more subtle than in the original model setting” – this is realistic, as real-world effects often attenuate compared to controlled trials. Even so, the direction of impact was consistently beneficial, underscoring the value of the intervention. In summary, the study’s findings are highly valuable: they bolster the case for geriatric-focused ED care models to enhance outcomes for an aging population, and they provide guidance (via analysis and lessons discussed) for other hospitals considering similar interventions.

Conclusion – Overall Confidence in the Results

After critically appraising Marsden et al. (2022) using the CASP cohort checklist, we can have a fairly high level of confidence in the study’s findings. The research question was clearly defined and the methodology sound for a cohort design. The inclusion of all eligible patients and adjustment for key confounders strengthens the validity of the results. While this was not a randomized trial (so we must consider possible residual confounding and time-related biases), the authors’ thorough analysis and consistent outcomes across two distinct hospitals make it likely that the benefits observed are truly attributable to the GEDI model rather than systematic bias. No major sources of bias (selection, measurement, or attrition bias) were detected in our appraisal – any limitations (like non-randomized allocation and contextual factors) were reasonably addressed by the study design or acknowledged in the interpretation.

Given that the study was conducted in Australia with a population and hospital context similar to those in Australia and New Zealand, the results appear applicable to geriatric emergency care in our region. In practice, this means we can be confident that implementing a program like GEDI in local EDs would likely lead to improved patient outcomes (e.g. more older patients managed without admission, shorter ED stays, fewer short-term returns, and potentially lower mortality) and possibly reduce costs of care for the elderly. In conclusion, Marsden et al.’s findings can be considered reliable and relevant – providing a strong evidence base with minimal bias for guiding improvements in emergency care for older adults in Australia and New Zealand . The overall risk of bias is low enough that health services can trust these results when planning geriatric interventions, and the evidence is directly pertinent to our local healthcare context.