Cost-effectiveness of a brief two-stage emergency department intervention for high-risk elders: results of a quasi-randomized controlled trial☆☆☆★

Introduction 

The emergency department is a critical point of contact with the health care system for older persons. It can also serve as a location to detect underserved or inappropriately served persons and to refer these persons to appropriate health care providers in the community. Various mechanisms have been used to rationalize the use of services to the elderly seen in the ED, including case-finding and liaison and case-management interventions. Several descriptive studies of ED-based case-finding and liaison interventions have been reported from different countries, including the United States1 and Canada.2, 3, 4, 5 Controlled studies of these interventions include 1 small randomized trial of a Canadian quick-response team6 and 2 nonrandomized trials of nursing assessment and referral interventions: the first of the latter trials did not find any health effect of an ED-based assessment,7 whereas the second, of a home-based assessment shortly after release from the ED, reported a beneficial effect on functional status.8 With regard to case management, one well-conducted randomized controlled trial of a nurse case-management intervention for older patients referred from the ED did not find any beneficial health effects.9 Unexpectedly, the intervention also resulted in an increase in ED use during the follow-up period. These mixed results suggest that more appropriate care of older ED patients can achieve better outcomes; costs can be more or less, depending on the nature of the services.

Only one of the aforementioned controlled intervention studies conducted an economic evaluation. A small randomized trial examined the costs and effects of ED-based home care coordinators, who provided prompt coordination of home care services for conditions that required increased intensity of services but were expected to resolve within 10 days.6 Although the study lacked power to detect clinically important effects, there appeared to be a shift in expenditures from hospital to home care services during the index ED visit and the 10-day follow-up period.

In this article, we report on a cost-effectiveness analysis of a quasi-randomized controlled trial of an ED-based, 2-stage, case-finding and liaison intervention for patients aged 65 years and older, consisting of screening with a 6-item questionnaire, followed by a brief standardized nursing assessment and referral to community providers. We have reported that the intervention was associated with a reduction in the rate of functional decline during the 4 months after the ED visit.10 The purpose of this report is to present new data on the societal costs of the intervention in comparison with the costs of usual care during the 4 months after the ED visit.

Materials and methods 

The study population was composed of individuals aged 65 years and older who visited 1 of 4 hospital EDs in Montreal, Quebec, Canada, and were expected to be discharged. Excluded were patients referred from a long-term care facility, those unable to communicate in English or French, nonresidents, patients who were medically unstable or cognitively impaired and had no family members as proxies, and patients who had received an ED consultation from a member of the hospital's geriatric staff before they were approached to be in the study. Informed consent was provided by the patient or, if cognitively impaired, by a family member. The study protocol was approved by the ethics review boards of the 4 hospitals.

In both treatment conditions, patients were administered a screening questionnaire (Identification of Seniors At Risk [ISAR]) by a research assistant comprising 6 self-report questions on functional dependence (premorbid and acute change), recent hospitalization, impaired memory and vision, and polymedication.11 The ISAR tool predicts functional decline and other adverse health outcomes,11 return ED visits,12 and high rates of hospital use.13 In the control group, high-risk patients (those with scores of ≥2 of a possible 6) received conventional care (consultation, treatment, and referral from the emergency physicians, with additional ED consultations as needed). The results of the ISAR screen for control group patients were not divulged to ED staff.

In the intervention group, high-risk patients received a brief standardized assessment by a study nurse, using a checklist developed by the authors. The assessment checklist, adapted from those used in other programs, included physical and mental function, medical status, and relevant social factors. The focus was on new problems and unresolved old problems that required either a medical intervention, a new or increased level of home-care or other service, or a comprehensive geriatric assessment; overall, 61.2% of patients assessed had one or more such problems. The average duration of the initial ED assessment was 28.0 (SD 12.57) minutes and ranged from 10 to 60 minutes. After completing the assessment, the study nurses consulted with hospital ED and geriatric staff and made referrals, as needed, to the local community health center, the primary physician, the geriatric outpatient clinic, or other community services. Most frequent referrals were to the primary physician (86.4%) and the local community health center (31.3%). In 38.1% of patients, there was limited follow-up by the intervention nurses, composed of telephone calls to the patient, the community health center, and/or the primary physician to help improve compliance with referrals.

Data sources included patient and caregiver questionnaires, intervention nurse records, hospital chart review, and provincial administrative databases (for community health center use, physician and medication billings, and hospital discharges). Patient outcomes assessed by telephone interview 4 months after enrollment included functional decline (using the same definition as in our original study11: a decline of ≥3 points on a 28-point activities of daily living scale14 or death) and change in depressive symptoms (15-item Geriatric Depression Scale).15 For cognitively impaired patients, proxies completed the activities of daily living instrument but not the depression measure.

Other measures made by the research assistant at the time of enrollment included demographic information (ie, age, sex, marital status, residence). Disability at baseline was a 4-category variable derived from the activities of daily living instrument: no disability, mild disability (in instrumental activities of daily living only), moderate disability (in 1 to 3 basic activities of daily living), and severe disability (in ≥4 basic activities of daily living). Patients completed a comorbidity checklist that has been validated against a chart-based comorbidity index.16 We used the 6-question Blessed Orientation-Memory-Concentration measure of cognitive impairment.17, 18 The weighted error score ranges from 0 to 26; the recommended cut-off point for cognitive impairment is a score of 7 or greater.17 The first 3 questions of this scale, on orientation to time and place, were used to decide whether to interview the patient or a proxy.

Several data sources on health services use were used, including provincial administrative databases, chart reviews, and questionnaire measures. Agreement between these sources was first examined; final study measures were derived primarily from administrative databases because these were the most complete for all study patients and generally agreed well with data from other sources. There was greater than 99% agreement between administrative databases and other sources on demographic variables, such as age (stratified in 5-year categories) and sex. There was 75% agreement between administrative databases and hospital medical charts on the number of outpatient clinic visits and 94% agreement on the number of ED visits. There was also a good agreement (85%) between the community health center database and patient's self-report.

Information available only from patient questionnaires, caregiver questionnaires, or both included use of non–community health center home-care and community services; personal costs not covered by Medicare (eg, for equipment, medications, travel to medical appointments); and nonprofessional home care provided by family and other unpaid caregivers.

All data were analyzed on an intention-to-treat basis. We used regression models to estimate the effect of the intervention on functional decline, change in depressive symptoms, and overall cost during the 4 months after the index ED visit, while simultaneously adjusting for the effect of patient characteristics. These models were estimated by using a Bayesian approach. Noninformative prior distributions were used for all parameters. We report the median and 95% credible intervals of the posterior distributions of the parameters of interest. These are the Bayesian equivalent of the classical point estimate and 95% confidence interval (CI). In fact, when noninformative prior distributions are used, the posterior credible intervals are numerically similar to the frequentist CIs. A Bayesian 95% credible interval can be interpreted as the interval that contains the parameter with 95% probability.

The effect of the intervention on functional decline or death, treated as a binary outcome, was estimated by using logistic regression. For the change in depressive symptoms and cost, we used linear regression models. For the model of the overall cost, we used a logarithmic transformation because of the nonnormal distribution of this variable. The ratio of the cost in the intervention group to the control group can be obtained as exp(β), where β is the regression coefficient corresponding to the intervention variable. Regression models for patient outcomes were adjusted for patient's age (<75 or ≥75 years), sex, ISAR score (2 or 3 to 6), cognitive impairment (no, yes, or missing), disability (mild or none versus moderate or severe), previous community health center use, comorbidity, residence (living at home alone or not), whether the patient had a caregiver, hospital of index ED visit, and triage category (ambulatory or stretcher). In models for patient health outcomes, we adjusted for previous community health center use measured from the self-report and for whether the patient interview was conducted after the 4-month follow-up period (>134 days from enrollment, 13% of follow-up interviews). In the model of overall cost, we adjusted for community health center use in the 30 days before the index date, ED use 30 days before the index date, and hospitalization 1 year before the index date, as measured from administrative data. In addition, we evaluated the effect of the interaction between the intervention group and 3 of the covariates: ISAR score, presence of a caregiver, and previous ED use. Descriptive statistics were calculated with SAS for Windows (version 6.12, SAS Institute, Inc., Cary, NC). The regression models were estimated with WinBUGS software (version 1.3, MRC Biostatistics Unit, Cambridge and Imperial College School of Medicine at St. Mary's Hospital, London, United Kingdom).19 For each model, sampler was assessed with the method of convergence of Gelman and Rubin.20 We ran 51,000 iterations and retained the last 50,000 to estimate the posterior distributions.

A cost-effectiveness analysis was conducted by using the guidelines set out by the Canadian Coordinating Office for Health Technology Assessment.21 Timelines for the study were from randomization until follow-up at 4 months. A societal perspective was taken, including ministry of health, patient, and caregiver resources. Multiple outcomes were assessed (functional decline and change in depressive symptoms) and analyzed separately according to the method of Bakker et al.22 Governmental costs included physician, hospital inpatient and outpatient, drug, and community health center costs. Private costs included patient and caregiver travel and out-of-pocket costs. The methods used to collect all data on services used and costs are summarized in the Appendix. A cost-effectiveness analysis was conducted comparing costs and outcomes for the 2 interventions. All costs are in 1999 Canadian dollars.

The sample size was calculated to estimate a 95% CI half-width of 6% for a difference of 10% or more in the functional decline rate between the intervention and control groups. The expected rate of functional decline in the control group was 30%.

Eligible patients were randomized by day of visit to the study intervention or to the control group. Each of 2 intervention nurses was assigned to 2 hospitals and rotated between them on a schedule assigned by the statistician using blocked randomization. High-risk patients recruited on intervention days were referred to the intervention nurse immediately after enrollment and completion of the baseline questionnaire. Patients recruited on control days received usual care. If an intervention nurse was absent on a particular day because of illness or other unplanned circumstances, patients were recruited into the control group.

Research assistants did not inform ED staff which patients were recruited into the study to maximize blinding of ED staff and to reduce contamination effects. However, the intervention nurses coordinated the intervention with other ED staff, who were therefore aware of certain intervention group patients. Follow-up interviews were conducted by telephone 4 months after the index ED visit by research assistants who were blind to the study group as far as possible (19% of follow-up interviews were conducted by the same research assistant who had enrolled the patient).

Results 

During the recruitment period, a total of 10,826 visits of patients aged 65 years and older were recorded, of which 7,921 were assessed for study eligibility criteria. A total of 5,755 of those assessed were excluded for the following reasons: patient expected to be admitted (n=2,781), already seen by geriatric team member (n=698), nonresident of Montreal (n=840), resident in nursing home (n=558), language barrier (n=265), previously enrolled in study (n=217), too sick and no proxy available (n=215), cognitive impairment and no proxy available (n=137), and death in the ED (n=34). Early in the study, 10 patients were inadvertently excluded because they had no primary caregiver. Of 2,166 eligible patients, 63 (2.9%) declined the screening, and 11 patients could not be found to complete the screening. Among 2,092 patients screened, 426 (20.1%) had positive results, with an ISAR score of 2 or more. A total of 388 (91.1%) of the 426 patients with positive ISAR scores consented to participate in the study; 178 were allocated to the intervention group and 210 to the control group.

At enrollment, the 2 groups were very similar with regard to demographic characteristics, health status, and prior health services use (Table 1).

Table 1. Patient characteristics by study group at enrollment.

	Characteristics	Intervention Group (N=178)	Control Group (N=210)
	Categoric variables, No. (%)
	Sex
	Female	99 (55.6)	137 (65.2)
	Male	79 (44.4)	73 (34.8)
	Triage category
	Ambulatory	104 (58.4)	126 (60.0)
	Stretcher	74 (41.6)	84 (40.0)
	Residence
	Home alone	62 (34.8)	95 (45.2)
	Other	116 (65.2)	115 (54.8)
	ISAR score
	2	100 (56.2)	108 (51.4)
	3	53 (29.8)	68 (32.4)
	4–5	25 (14.0)	34 (16.2)
	Information score
	Patient	158 (88.8)	186 (88.6)
	Proxy	20 (11.2)	24 (11.4)
	Family caregiver
	No	42 (23.6)	73 (34.8)
	Yes	136 (76.4)	137 (65.2)
	Cognitive impairment
	No	97 (61.4)	110 (59.8)
	Yes	61 (38.6)	74 (40.2)
	Missing	20	20
	Disability (premorbid)
	None–mild	93 (52.3)	110 (52.4)
	Moderate–severe	85 (47.7)	100 (47.6)
	ED visit 1 month before index visit	55 (31.1)	71 (34.0)
	Missing	2	2
	Primary physician visit during month before index visit	76 (43.2)	86 (41.4)
	Missing	2	2
	Hospitalization during year before index visit	91 (51.1)	114 (54.3)
	Face-to-face CHC visit during month before index visit	34 (19.9)	54 (26.1)
	Missing	7	3
	Continuous variables, No. (Mean [SD])
	Age	178 (76.7 [7.1])	210 (76.5 [7.0])
	Depressive symptoms*	150 (4.9 [3.5])	177 (4.6 [3.2])
	Comorbidity score*	178 (2.4 [2.2])	210 (2.3 [1.9])
	Length of stay in ED, h	153 (12.6 [14.0])	169 (12.2 [15.0])
	Median	6.8	6.0
	IQR	3.7–19.4	3.3–15.6
	*Higher score indicates greater impairment or symptoms.

CHC, Community health center; IQR, interquartile range.

Health outcomes for the 2 study groups are shown in Tables 2 and 3.

Table 3. Change in patient depressive symptoms 4 months after index visit.

	Outcome	No. of Patients		Mean (SD)		Regression Coefficient: Posterior Median (95% Credible Interval)
		Intervention Group	Control Group	Intervention Group	Control Group	Unadjusted Model	Adjusted Model*
	Change in depressive symptoms†	103	125	−0.4 (3.0)	0.1 (3.0)	−0.5 (−1.3 to 0.3)	−0.5 (−1.3 to 0.3)
	*Models for patient outcomes were adjusted for age, sex, ISAR score, cognitive impairment, disability, previous community health center use, comorbidity, residence, caregiver involvement, hospital of index ED visit, and triage category. †Difference in measurements at 4 months relative to baseline. A negative change score indicates improvement.

Table 2. Decline in patient functional status 4 months after index visit.

	Outcome	No. of Patients		% of Patients With Outcome		OR: Posterior Median (95% Credible Interval)
		Intervention Group	Control Group	Intervention Group	Control Group	Unadjusted Model	Adjusted Model*
	Decline in patient functional status or death (4 mo)†	152	178	21.1	30.9	0.6 (0.4 to 1.0)	0.5 (0.3 to 0.9)
	*Models for patient outcomes were adjusted for age, sex, ISAR score, cognitive impairment, disability, previous community health center use, comorbidity, residence, caregiver involvement, hospital of index ED visit, and triage category. †Number of deaths: 5 in intervention group and 8 in control group.

We successfully linked all 388 patients with the hospital discharge database, 386 patients with the physician and prescription billing database, and 378 patients with the community health center database. Measures of health care use and costs, showing costs in patients with complete data and costs for all available data, respectively, are shown in Tables 4 and 5.

Table 4. Health care costs by study group during the 4 months after the index visit among patients with complete information from all sources.

	Resource Item	No. of Billings-Contacts				Cost per Patient (Canadian $)
		Intervention Group (N=125)		Control Group (N=140)		Intervention Group (N=125)		Control Group (N=140)		Difference (95% CI)
		Mean	SD	Mean	SD	Mean	SD	Mean	SD
	Public resources: Total	—	—	—	—	3,193	4,092	3,574	4,057	−382 (−1,369 to 605)
	Study nurse intervention	1.21	0.38	0.00	0.00	30	10	0	0
	Physician visits in ED	3.66	4.74	3.85	6.04	128	166	131	206	−2 (−48 to 43)
	Physician visits in OPD or acute hospital	11.91	18.13	14.36	20.93	385	613	545	913	−160 (−351 to 30)
	Physician visits in others places	5.50	7.64	5.62	12.16	143	188	149	239	−6 (−58 to 47)
	OPD visits (hospital component)	4.02	5.36	3.94	5.22	—	—	—	—
	ED visits (hospital component)	0.98	1.21	0.91	1.33	91	112	84	123	7 (−22 to 36)
	Prescription drugs	26.90	22.53	25.81	21.80	934	887	769	584	165 (−15 to 345)
	Acute hospitalization*	0.30	0.56	0.44	0.72	1,154	3,079	1,439	2,898	−285 (−1,008 to 439)
	Admissions to psychiatry-rehabilitation	0.02	0.13	0.04	0.20	2	13	5	23	−3 (−8 to 2)
	Day surgery	0.06	0.35	0.04	0.20	8	46	16	100	−8 (−27 to 12)
	CHC—telephone contacts	2.41	6.00	4.41	20.65	6	16	12	55	−5 (−16 to 5)
	CHC—clinic visits	1.86	7.33	2.26	7.40	107	415	128	422	−21 (−123 to 80)
	CHC—home/other visits	3.82	11.16	5.56	12.53	204	554	297	649	−94 (−240 to 53)
	Private resources: Total	—	—	—	—	544	653	549	718	−5 (−172 to 162)
	Out-of-pocket patient	—	—	—	—	251	335	250	516	1 (−106 to 108)
	Out-of-pocket caregiver	—	—	—	—	56	133	43	181	13 (−26 to 52)
	Travel (trips) patient	8.72	17.26	7.01	14.35	87	173	70	143	17 (−21 to 55)
	Travel (trips) caregiver	3.90	10.79	1.95	5.48	39	108	20	55	19 (−1 to 40)
	Non-CHC services	2.86	6.44	4.00	10.56	111	239	166	440	−55 (−143 to 32)
	Grand total	—	—	—	—	3,737	4,137	4,124	4,306	−387 (−1,411 to 638)
	*Excluding admissions to psychiatry-rehabilitation.

OPD, Outpatient department.

Table 5. Health care costs by study group during the 4 months after the index visit by using all available data.

	Resource Item	No. of Patients		No. of Billings-Contacts				Cost per Patient (Canadian $)
				Intervention Group		Control Group		Intervention Group		Control Group		Difference (95% CI)
		Intervention Group	Control Group	Mean	SD	Mean	SD	Mean	SD	Mean	SD
	Public resources
	Study nurse intervention	178	210	1.19	0.40	0.00	0.00	30	10	0.00	0.00
	Physician visits in ED	177	209	3.64	4.50	3.96	6.09	128	160	140	224	−12 (−52 to 28)
	Physician visits in OPD or acute hospital	177	209	15.22	25.64	16.45	25.40	536	1,239	637	1,119	−101 (−337 to 135)
	Physician visits in others places	177	209	5.16	6.93	5.37	10.38	141	186	147	215	−6 (−47 to 34)
	ED visits (hospital component)	177	209	1.01	1.14	0.94	1.31	93	106	87	121	6 (−17 to 29)
	OPD visits (hospital component)	177	209	3.80	5.08	3.88	5.54	—	—	—	—
	Prescription drugs	178	210	26.03	26.07	25.90	25.21	868	817	795	874	73 (−97 to 243)
	Acute hospitalization*	178	210	0.40	0.72	0.53	0.79	1,742	4,056	2,185	4,434	−444 (−1,298 to 411)
	Admissions to psychiatry-rehabilitation	178	210	0.02	0.15	0.04	0.20	2	15	5	23	−3 (−7 to 2)
	Day surgery	178	210	0.07	0.35	0.04	0.23	12	76	15	99	−3 (−21 to 15)
	CHC—telephone contacts	171	207	2.77	6.72	3.89	17.34	7	18	10	46	−3 (−10 to 4)
	CHC—clinic visits	171	207	1.76	6.44	2.47	7.46	106	374	144	434	−39 (−122 to 45)
	CHC—home/other visits	171	207	3.52	10.21	6.30	16.01	190	513	339	836	−149 (−293 to −5)
	Private resources
	Out-of-pocket patient	—	—	—	—	—	—	235	315	240	473	−4 (−95 to 86)
	Out-of-pocket caregiver	—	—	—	—	—	—	51	126	40	169	11 (−23 to 45)
	Travel (trips) patient	147	172	7.95	16.20	6.53	13.17	79	162	65	132	14 (−18 to 47)
	Travel (trips) caregiver	143	162	7.04	43.56	1.75	5.13	70	436	17	51	53 (−15 to 121)
	Non-CHC services	146	172	3.20	8.34	3.75	9.82	126	320	154	406	−28 (−109 to 54)
	*Excluding admissions to psychiatry-rehabilitation.

Table 4, in which costs can be summed, indicates that overall costs were lower in the intervention group than in the control group by $387 per patient, a 9.4% reduction relative to the control group. The 95% CI for this difference, however, does not lie outside the range of equivalence (ie, it includes 0). Furthermore, both Table 4, Table 5 indicate that differences in costs in individual categories are mostly small, and that in most cases, their 95% CIs do not lie outside the range of equivalence. The difference in costs is largest by far for acute hospitalization ($285 in Table 4) and favors the intervention group, although again the 95% CI does not lie outside the range of equivalence. In only a few cost categories is the intervention more expensive. The most important such difference is that for prescription drugs, where the average cost per patient is higher by $165 in the intervention group (Table 4). Again, however, the 95% CI includes 0. The intervention cost itself is negligible ($30 per patient, or <1% of the total cost). Similar results were obtained by using all available data (Table 5). Costs were not available for the hospital component of outpatient clinic visits.

Not included in Table 4, Table 5 are data on nonprofessional costs. Among participating caregivers, there was no important difference in the mean number of hours per week spent caring for the patients in the intervention group versus those in the control group, either by the primary caregiver (mean of 14.3 [SD 23.9] versus mean of 12.5 [SD 15.0], respectively) or by other caregivers (mean of 2.3 [SD 5.2] versus mean of 2.6 [SD 4.8], respectively).

From an unadjusted model for the log (cost), we found that the posterior median of the ratio of the cost in the intervention to the control group was 0.92 (95% credible interval 0.73 to 1.16). After adjusting for the effect of covariates, it increased marginally to 0.94 (95% credible interval 0.75 to 1.17). The adjusted model is presented in Table 6.

Table 6. Results of multiple linear regression analysis for overall cost (n=265).

	Independent Variables	Regression Coefficient		Ratio of Costs*
		Posterior Median	Posterior 95% Credible Interval	Posterior Median	Posterior 95% Credible Interval
	Experimental group
	Control†	0.00		1.00
	Intervention	−0.06	−0.28 to 0.16	0.94	0.75 to 1.17
	Age, y
	<75†	0.00		1.00
	≥75	−0.15	−0.38 to 0.09	0.86	0.69 to 1.09
	Sex
	Female†	0.00		1.00
	Male	−0.09	−0.33 to 0.15	0.91	0.72 to 1.15
	Residence
	Home alone†	0.00		1.00
	Other	0.06	−0.18 to 0.31	1.06	0.83 to 1.36
	ISAR score
	2†	0.00		1.00
	3–6	0.13	−0.09 to 0.37	1.14	0.91 to 1.45
	Cognitive impairment
	No†	0.00		1.00
	Yes	−0.05	−0.30 to 0.19	0.95	0.74 to 1.22
	Missing	−0.15	−0.50 to 0.18	0.86	0.61 to 1.21
	Disability
	Mild or none†	0.00		1.00
	Moderate or severe	0.13	−0.10 to 0.37	1.14	0.91 to 1.45
	Comorbidity	0.10	0.05 to 0.16	1.11	1.05 to 1.17
	Family caregiver
	No†	0.00		1.00
	Yes	0.03	−0.23 to 0.28	1.03	0.79 to 1.32
	Hospital of index visit
	Hospital 1†	0.00		1.00
	Hospital 2	−0.22	−0.49 to 0.06	0.80	0.61 to 1.06
	Hospital 3	−0.24	−0.65 to 0.17	0.79	0.52 to 1.18
	Hospital 4	−0.08	−0.38 to 0.23	0.93	0.68 to 1.26
	Triage category
	Ambulatory†	0.00		1.00
	Gurney	0.11	−0.12 to 0.34	1.12	0.89 to 1.41
	CHC use in previous 30 d
	No†	0.00		1.00
	Yes	0.45	0.18 to 0.72	1.57	1.20 to 2.06
	Hospitalized in the previous year
	No†	0.00		1.00
	Yes	0.24	0.02 to 0.47	1.28	1.02 to 1.59
	ED use in previous 30 d
	No†	0.00		1.00
	Yes	0.16	−0.09 to 0.40	1.17	0.92 to 1.49
	*For a continuous variable, the ratio is between the cost at a given value of the variable compared with one unit lower. For categoric variables, the ratio is between the cost at each value of the variable and the reference category. †Reference category.

Comparing the net effects of the intervention on costs and outcomes, we note that there was a small effect, favoring the intervention group, in government or societal costs, with a 95% credible interval including a zero difference. In the context of our study, because the outcomes are superior in the intervention group, virtually any cost savings signifies dominance (and therefore greater cost-effectiveness) of the intervention.

Discussion 

This study contributes to the growing literature on economic evaluations of health care interventions for older persons, including studies of outpatient interventions,23, 24, 25 assessment and referral interventions for unselected members of the general population,26, 27 posthospital discharge interventions,28, 29 inpatient interventions,30, 31, 32, 33 and case management programs for frail elderly persons living in the community.34, 35, 36, 37 Because inpatient hospital costs are generally the single highest cost item (as in our study), those interventions that reduced overall costs did so usually by reducing inpatient hospital costs.28, 29, 37, 38, 39, 40 However, the cost data reported in these studies were generally limited to reimbursements for insured services; they did not include private costs. In contrast to these studies, we attempted to conduct a comprehensive assessment of overall societal costs. Because many health care services are publicly funded in Quebec, including physicians, medication prescriptions, and most medically necessary home-care and community-based services, we were able to capture the great majority of societal costs from the provincial administrative databases. The proportion of total mean costs per patient that were from public resources was 85.4% and 86.6% in the intervention and control groups, respectively. Even so, we excluded some potentially important costs, particularly the costs of nonprofessional care. However, we found no important difference in the average weekly hours of care in the 2 study groups.

Although the overall estimate of the effect of intervention indicated only a small cost savings, in the subgroup of patients who had visited the ED during the month before the index date, the cost savings were more substantial, estimated at 0.66 of those in the control group, with a 95% CI that excluded equivalence. Although this secondary finding must be interpreted with caution, it suggests the potential for cost savings among patients with higher-than-average expected costs.

The internal validity of this study might be reduced by bias and imprecision. Two potentially important sources of bias should be considered. First, we did not randomize individuals but rather days. However, this quasirandomization was the most feasible method and resulted in 2 study groups that were quite similar at baseline; any potentially important differences between the study groups were adjusted for in the multivariate analysis. Second, the effectiveness of the intervention might have been underestimated because of (1) the exclusion of patients who had received a geriatric consultation before study enrollment (ie, patients who might have stood to benefit the most from the intervention); (2) possible contamination of the control group (eg, provision of services similar to those received by the intervention group) because of our inability to completely blind ED staff as to who received the intervention; and (3) failure to deliver the intervention to 17% of patients in the intervention group, presumably resulting in a dilution of the intervention effect. Imprecision in the measurement of some variables, such as those that relied on questionnaires to assess personal costs, probably increased the variability in some of the outcomes measured.

The external validity, or generalizability, of the results is uncertain. Characteristics of the context of the study, including an urban versus rural setting, the organization and access to health services, and the presence of a government-funded national health care system, are likely to play a major role in the effectiveness of the intervention. In the present study, for example, patients encountered delays in obtaining home care services, particularly if they were not previous users, and difficulties in accessing a primary care physician unless they already had an ongoing relationship with their physician (McCusker et al, unpublished data, 2002). Such problems might have reduced the apparent effectiveness of the intervention.

The time frame for measurement of outcomes might not have been long enough to determine the effect of the intervention on costs. One study of a geriatric outpatient intervention that did not find any cost reduction during the first 12 months of follow-up24 later reported cost savings during the 16- to 24-month period, suggesting that longer follow-up is needed to observe investment effects of these interventions.41

This project has highlighted the pivotal role of the ED in the interface between hospital and community services. Seniors visiting the ED are at high risk of functional decline and incur high rates of health care use and costs during the months after the ED visit. This project has demonstrated the critical role played by liaison staff in ensuring the continuity of care between hospital-based and community-based services. It has also demonstrated that the efficiency of delivery of these liaison services can be improved. Currently, in most EDs that have such liaison staff (usually nurses or social workers), referrals to these staff are made on an ad hoc basis and are highly dependent on volume. A systematic approach with tools for the identification of high-risk patients and rapid assessment of those at high risk has the potential to ensure the quality and consistency of liaison interventions, even in the face of rising patient volumes. The approach tested in this study can help improve the process of care in the ED for those discharged to the community. By providing key information on the risk of functional decline to partners in the community, the rate of further functional decline can be reduced at no greater societal cost than is involved in providing usual care.

In conclusion, the results of this study indicate that a brief 2-stage ED nursing intervention for older persons who were expected to be released from the ED did not increase overall societal costs, considering public and private costs incurred either at or during the 4 months after the ED visit. Thus, with regard to cost-effectiveness, the study intervention is preferred because at least one beneficial outcome was observed (a reduced rate of functional decline), there were no adverse outcomes, and overall societal costs appeared slightly lower, or at least no higher, than if usual care only were given.