Case finding and referral model for emergency department elders: A randomized clinical trial☆☆☆★

Introduction 

The emergency department is a common entry point into the health care system for many ill elderly patients and has been increasingly used by this population over the past decade.1, 2 Up to 18% of those who use the ED are 65 years and older,2, 3, 4 and those 75 years and older have the highest rates compared with all other age groups.2 ED visits by older adults are often medically necessary and typically herald increased frailty, decline, and likelihood of institutionalization.1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15

Older ED patients have complex medical, social, and physical problems that might be amenable to coordinated care or case management. Few programs or models of care that provide comprehensive assessment for the complex, multiple needs of geriatric patients in the ED setting have been reported.16, 17, 18, 19 The only controlled trial examining the effectiveness of a comprehensive geriatric assessment program on elderly outcomes demonstrated negative results. The investigators postulated that the lack of effect on outcomes might have been a function of inadequate sample size, nonrandomized design, or lack of direct intervention for medical and community services on discharge from the ED.17

A number of case management or care-coordinated programs have been implemented successfully with hospitalized elderly patients.20, 21, 22, 23, 24, 25, 26 One model not dependent on a specific disease category, the transitional model of care, has been shown to be effective for hospitalized older patients.27, 28 The transitional model uses a comprehensive discharge planning protocol implemented by advanced practice nurses and short-term home follow-up by the advanced practice nurse in lieu of a visiting home nurse. Naylor et al27, 28 devised the transitional model on the premise that timely and comprehensive assessment of older hospitalized patients of needed home services would reduce subsequent costly service use (ie, hospital readmissions and ED visits). They have demonstrated that the model reduced rates of hospital readmission, increased the time from the index hospital discharge to readmission, and lowered health care costs over the ensuing 24 weeks compared with those of older hospitalized patients receiving usual discharge and follow-up community services.28

The hospital transitional model of care might be applicable to older adults in the ED because many elderly ED patients have multiple comorbid illnesses and concomitant psychosocial needs. We theorized that identifying unmet medical, health, and social needs of older ED patients and linking them to appropriate follow-up medical care, social, or home health care services would delay functional decline and subsequent nursing home admission, reduce repeat ED visits, and lower hospital rates. Thus, we established a formalized linkage and referral system between an ED-based advanced practice nurse and 10 public and not-for-profit social and health community agencies.29 The objective of this randomized clinical trial was to examine the effectiveness of this model of care for community-dwelling older ED patients. We hypothesized that this intervention would result in lower 30- and 120-day rates of service use, defined as subsequent ED visits, hospitalizations, or nursing home admissions; lower health care costs; and higher rates of community referrals among the intervention group compared with those in the usual care group. In addition, we hypothesized that the intervention group would have greater satisfaction with ED nursing care and less decline in physical and mental health at 30 and 120 days compared with those individuals in the usual care group.

Materials and methods 

A randomized clinical trial was conducted at 2 EDs of urban academic teaching centers. Hospital 1, a 1,000-bed teaching hospital, is located in a neighborhood federally designated as a medically underserved community. Hospital 1's ED consists of 18 beds and has more than 40,000 visits annually. Hospital 2 is a 700-bed county teaching hospital that has a 35-bed ED and is a Level I trauma center. Hospital 2 receives more than 70,000 ED visits annually and serves a primarily indigent population. The study was reviewed and approved by the institutional review boards at both institutions.

Potential participants were 65 years or older, resided in the community, and visited 1 of the study ED sites from October 1, 1999, through May 15, 2000 (Hospital 2), or June 16, 2000 (Hospital 1; the extra month was added to increase sample size). Participants were eligible if they were expected to be discharged home from the ED; resided within the geographic service area; had telephone access; were able to hear, understand, and speak English at the time of the ED visit; and had a primary family caregiver willing to serve as a proxy respondent in cases of significant cognitive impairment (defined as ≥5 errors on the Short Portable Mental Status Questionnaire).30

There were 7,017 elderly individuals who made 9,240 ED visits during the study period. The number of ED visits per individual ranged from 1 to 18 visits, with 1,431 (20%) having more than 1 ED visit. Of the 7,017 older adults, 2,815 (40%) were screened for study eligibility; screenings were conducted on the basis of research personnel availability. Of those screened, 987 (35%) were eligible; of those eligible, 650 (66%) were enrolled in the study.

Individuals screened and not screened were similar in terms of age, sex, and race. Screened individuals were more likely to be assigned a triage classification of “emergency” or “urgent” than nonscreened individuals (34.1% versus 28.9%). Enrollees (n=650) and those who were eligible but declined to participate (n=337) were similar in terms of age, sex, race, and triage classification.

A randomized clinical trial was conducted by using block randomization after assessment of risk status (Figure)

View Large Image Download to PowerPoint

Figure. Flow diagram of study participants.

Trained research assistants screened potential participants for eligibility and enrolled participants during 5 shifts a week; 4 of the shifts were during weekdays between the hours of 8 AM and 5 PM , and 1 evening or weekend shift was chosen at random. After determining risk status (low or high) from the nurses' triage assessment, participants were subsequently randomized to the intervention group or the usual care group within each risk status by using a block randomization scheme. The scheme was designed by a biostatistician to ensure blinded assignment by the research assistants and equivalent numbers of participants by risk status. If the participant was assigned to the intervention group, the research assistant notified the advanced practice nurse, who was kept unaware of the triage nurses' risk status determination.

Sample size determination was based on the estimate of a 30% rate of service use within 4 months of the index ED visit for the usual care group on the basis of previous literature review and data on 1-month repeat ED use at the 2 study sites. A 20% rate of service use was projected for the intervention group (an absolute reduction of 10%). By using a 2-tailed χ2 test comparing proportions with an α value of .05 and a power of 0.90, our targeted sample to detect a 10% difference in repeat ED visits was 400 per group or a total of 800 patients. Sample size determination was based solely on the 2 groups; the block design was developed to evaluate the predictive ability of the triage risk-screening instrument.31 We had no a priori data to determine whether the designated high-risk and low-risk groups would differ in terms of outcome rates.

All study interviews (at baseline and 30 and 120 days after the index ED visit) were conducted by research assistants who had no role in the intervention. One set of research assistants conducted the baseline (index ED visit) data collection, and a separate set of trained interviewers conducted the follow-up interviews. Those conducting the follow-up assessments were blinded to group assignment and hypotheses of the study. All research assistants underwent training and followed standard procedures outlined in a training and coding manual. All data were collected on standardized, precoded forms; interrater reliability measurements verified the consistency of ratings among all research assistants (κ=0.90 on all items). Ongoing quality checks were conducted throughout the course of the study. All data were entered twice into a computerized database and underwent extensive checks of error and validity.

After obtaining informed consent, baseline data were collected by using a structured questionnaire. The baseline interview included the Short Portable Mental Status Questionnaire,30 the MOS 36-Item Short-Form Health Survey (SF-36),32 additional questions on ability to perform activities of daily living and instrumental activities of daily living, and questions on current use of community services. Research assistants abstracted the following information from the medical record: nursing assessment of risk status; triage level (1 [emergency] to 4 [simple]); reason for the ED visit; number of medications used; use of ED and hospital 30 days before the ED visit; and evidence of a referral to a community agency or primary care provider on discharge.

Participants were interviewed by telephone by trained interviewers at 30 and 120 days after the index ED visit. Information was gathered on satisfaction with ED care (first month only); activities of daily living and instrumental activities of daily living function; use of EDs, hospitals, or nursing homes; use and knowledge of community services; and health-related quality of life by using the SF-36.32 Finally, financial and administrative databases at each hospital were examined to gather information on ED use, hospitalization, and health care charges.

Usual care participants received usual and customary ED care. ED recommendations for follow-up care or use of community agencies were given to participants or their proxies. Participants were responsible for contacting the suggested community agency or primary care provider whenever ED personnel recommended use of community services.

In addition to the usual ED care, participants in the intervention group received a comprehensive geriatric assessment during the index ED visit by an advanced practice nurse specializing in geriatrics. Information gathered from the standardized assessment29 and clinical judgment were used to identify unmet medical, social, or health care needs that could be addressed by a community or social service agency. The assessment focused on the participant's living situation and arrangements, physical function, cognitive status, nutritional status, depressive symptomatology, current comorbid conditions, alcohol use, potential for elder abuse, and current use of home-care services and agencies delivering the services. If the primary informal caregiver was present in the ED, he or she was also assessed to determine his or her knowledge, skills, and stress level. On the basis of the information from the assessment, the advanced practice nurse collaborated with the participant (and caregiver if available), the emergency physician, nursing personnel, and an ED social worker to design a discharge plan to address unmet needs. Participants were given a choice of agencies; if they agreed, a referral was faxed from the ED to the community agency to expedite home care services. Discharge summaries were sent to the participant's primary care provider detailing the ED visit, the reason for the visit, a summary of the advanced practice nurse assessment findings, and discharge plans from the ED.

The advanced practice nurses assumed responsibility for discharge planning while the patient was in the ED, with short-term telephone follow-up as necessary until community agency personnel achieved contact with the participant. The same advanced practice nurses conducted the intervention throughout the study period, with the exception of an additional advanced practice nurse at hospital 1 substituting during the study advanced practice nurse's vacation.

The primary end points were service use defined as subsequent ED visit, hospitalization, or nursing home admission at 30 and 120 days after the index ED visit. Data on subsequent use of the ED, hospitalization, and death were obtained through medical record abstraction at the 2 study sites and through telephone interviews. A positive finding by means of either method was considered an outcome event. Nursing home admission was obtained through the follow-up telephone interviews.

Health care costs of subsequent ED use and hospitalization after the index ED visit were assigned on the basis of Medicare reimbursement rates. Specifically, the cost of each ED visit was equal to the Medicare hospital outpatient reimbursement for the Ambulatory Payment Classification associated with the type of episode. Inpatient costs were estimated by using the median Medicare reimbursement for the coded Diagnosis-Related Group. All costs were expressed in US dollars for the year 2000. Costs associated with the index ED visit were not included in the 30- and 120-day cost estimates.

Changes from baseline in the 2 major subscale components of Physical Health and Mental Health of the SF-3632 were examined at 30- and 120-day follow-up. Satisfaction with nursing care was measured by using an existing satisfaction instrument already in use in the EDs that has 5 questions related to nursing care. A sixth item was added: “How would you rate your satisfaction with the information nurses or social workers gave about agencies or organizations that could help with your needs after you left the emergency room?” All questions had 5-point Likert responses ranging from poor (1) to excellent (5); each item was treated as a continuous variable.

The effectiveness of the intervention was examined by using an intention-to-treat approach. Additional analyses of effectiveness were conducted by excluding participants admitted to the hospital after study enrollment instead of returning home as initially planned.

Thirty-day and 120-day outcomes were examined between the 2 groups by using χ2 likelihood ratio tests for categoric variables, t tests for approximately normally distributed variables, and Wilcoxon rank-sum tests for skewed variables. Subgroup analyses were conducted, comparing the intervention and usual care groups by risk status. All statistical estimates are shown as estimates (95% confidence interval [CI]). Effects were considered statistically significant if CIs for odds ratios (ORs) failed to span 0.0. All analyses were performed by using SAS software (version 8.0, SAS Institute, Inc., Cary, NC).

Results 

A total of 650 participants enrolled in the study: 326 were randomized to the intervention group and 324 to the usual care group. The mean age of the entire sample was 74 years; 59% were women, and 39% were white. Participant characteristics at the time of the index ED visit are shown in Table 1.

Table 1. Participant characteristics at the index ED visit, according to study group (n=650).

	Characteristic	Intervention Group (N=326)	Usual Care Group (N=324)
	High-risk status, No. (%)	146 (45)	145 (45)
	Age, y±SD	74.4±6.5	74.5±7.3
	Female sex, No. (%)	180 (55)	205 (63)
	White race, No. (%)	133 (41)	120 (37)
	Education, y ±SD	11.7±3.8	10.9±3.6
	Medicaid, No. (%)	47 (14)	47 (15)
	SPMSQ
	Score, mean±SD	8.1±1.9	8.0±1.8
	5 errors, No. (%)	6 (2)	12 (4)
	SF-36, mean ±SD
	Physical Component	37.1±12.2	36.9±12.1
	Mental Component	51.6±11.3	51.9±10.3
	Current comorbidities, mean ±SD	3.06±2.84	3.08±1.97
	Current medications, mean ±SD	4.40±3.18	4.56±3.42
	Requires help with, No. (%)
	Bathing	55 (17)	49 (15)
	Dressing	35 (11)	49 (15)
	Getting in and out of bed/chair	37 (11)	30 (9)
	Eating	25 (8)	23 (7)
	Using the telephone	35 (11)	33 (10)
	Shopping	112 (34)	116 (36)
	Transportation	98 (30)	95 (29)
	Taking medications	49 (15)	53 (16)
	Banking	64 (20)	61 (19)
	Preparing meals	66 (20)	65 (20)
	Doing laundry	95 (29)	80 (25)
	Nurses' triage level, mean ±SD	3.13±0.72	3.06±0.74
	Emergency, No. (%)	1 (<1)	4 (1)
	Urgent, No. (%)	60 (19)	63 (21)
	Semiurgent, No. (%)	146 (47)	149 (49)
	Nonurgent, No. (%)	102 (33)	88 (29)
	Arrived to ED by ambulance or EMS, No. (%)	53 (16)	49 (15)

SPMSQ, Short Portable Mental Status Questionnaire.

Table 2 displays the overall service use and subsequent ED visits, hospitalizations, and nursing home admissions at 30 and 120 days after the ED index visit. Overall service use was similar between the 2 groups at 30 and 120 days after the ED index visit. Only nursing home admission within 30 days of the index ED visit differed significantly between the 2 groups, with intervention participants less likely to go to a nursing home (0.7% versus 3.0%; OR 0.21; 95% CI 0.05 to 0.99; Table 2).

Table 2. Subsequent ED visits, hospitalizations, and nursing home admissions at 30* and 120 days† after the index ED visit by group.

	Variable	Intervention Group, No. (%) (N=326)	Usual Care Group, No. (%) (N=324)	OR	95% CI
	30-d service use ‡	90 (28)	82 (25)	1.13	0.79 to 1.60
	Repeat ED visit	66 (20)	49 (15)	1.42	0.95 to 2.14
	Hospitalization	46 (14)	46 (14)	0.99	0.64 to 1.54
	Nursing home§	2 (0.7)	9 (3)	0.21	0.05 to 0.99
	Died	4 (0.6)	2 (0.3)	2.00	0.36 to 11.00
	120-d service use	157 (48)	159 (49)	0.96	0.71 to 1.31
	Repeat ED visit	121 (37)	128 (40)	0.90	0.66 to 1.24
	Hospitalization	91 (28)	87 (27)	1.05	0.75 to 1.49
	Nursing home§	5 (2)	12 (4)	0.40	0.14 to 1.15
	Died	9 (1)	10 (2)	0.89	0.36 to 2.72
	*Thirty-day results are based on all participants' medical records and telephone interviews of 287 (88%) participants in the intervention group and 281 (87%) participants in the usual care group. †One hundred twenty–day results are based on all participants' medical records and telephone interviews of 272 (83%) participants in the intervention group and 269 (83%) participants in the usual care group. ‡Service use refers to the presence of subsequent ED visits, hospitalizations, or nursing home admissions. §Nursing home results are based on telephone interviews only.

The total number of hospital days after the index ED visit was not distributed differently between the 2 groups for 30- and 120-day comparisons, respectively (Table 3). The number of hospital days was tested between groups for just those participants who had a subsequent hospitalization by using a t test on the natural logarithm of hospital days; the intervention did not affect hospital days at 30 or 120 days among those hospitalized.

Table 3. Number of hospital days and health care costs subsequent to the index ED visit at 30 and 120 days.

	Variable	30 Days After Index ED Visit			120 Days After Index ED Visit
		Intervention Group (N=326)	Usual Care Group (N=324)	Difference (95% CI)	Intervention Group (N=326)	Usual Care Group (N=324)	Difference (95% CI)
	Total No. of hospital days
	Mean (SD)	0.36 (±1.78)	0.76 (±3.85)	0.4 (−0.1 to 0.9)	1.37 (±5.11)	1.69 (±6.22)	0.3 (−0.6 to 1.2)
	Median	0	0	0 (0 to 0)	0	0	0 (0 to 0)
	Range	0–21	0–46		0–67	0–48
	Costs for subsequent ED visits, $
	Mean (SD)	19 (±55)	13 (±44)	6 (−1 to 14)	49 (±111)	39 (±80)	10 (−5 to 25)
	Median	0	0	0 (0 to 0)	0	0	0 (0 to 0)
	Range	0–412	0–309		0–928	0–424
	Costs for hospitalizations, $
	Mean (SD)	501 (±2,349)	643 (±2,333)	−142 (−502 to 219)	1,592 (±4,809)	1,865 (±6,831)	−272 (−1,182 to 637)
	Median	0	0	0 (0 to 0)	0	0	0 (0 to 0)
	Range	0–21,944	0–16,230		0–37,245	0–90,468

Health care costs for subsequent ED visits did not differ significantly between the 2 groups at either 30 or 120 days (Table 3). Analyses of the ED costs for only the participants who used the ED within the follow-up period showed no important difference between the 2 groups at 30 or 120 days after the index ED visit.

Health care costs for hospital admissions did not differ significantly between the 2 groups at either 30- or 120-day follow-up (Table 3). Mean cost was also compared between groups for only those participants who had one or more hospital admissions (using t tests on logarithm of cost). Again, there were no important differences between groups at 30 or 120 days.

Participants' self-reports of ED and hospital use agreed with the hospital-based data source in 89% of cases. Reanalysis of subsequent ED use or hospital admission using only the hospital-based data demonstrated no important differences between the intervention and usual care groups. Community agency referrals were significantly greater among the intervention group compared with those among the usual care group (56% versus 1%; OR 59.3; 95% CI 21.6 to 163.1).

Subgroup analyses were conducted by risk classification per the nurses' triage screen (Table 4).

Table 4. Subsequent ED visits, hospitalizations, or nursing home admissions among low- and high-risk participants by study group at 30 and 120 days after the index ED visit (n=650).*

	Variable	Low-Risk Group (N=359)			High-Risk Group (N=291)
		Intervention Group (N=180)	Usual Care Group (N=179)	OR or Mean Difference (95% CI)	Intervention Group (N=146)	Usual Care Group (N=145)	OR or Mean Difference (95% CI)
	30-d service use †(%)	38 (21)	23 (13)	1.8 (1.0 to 3.2)	52 (36)	59 (41)	0.8 (0.5 to 1.3)
	Subsequent ED use	31 (17)	18 (10)	1.9 (1.0 to 3.5)	35 (24)	31 (21)	1.2 (0.7 to 2.0)
	Hospitalization	16 (9)	9 (5)	1.8 (0.8 to 4.3)	30 (21)	37 (26)	0.8 (0.4 to 1.3)
	Total hospital days (±SD)	0.2 (±1.2)	0.1 (±0.7)	0.1 (−0.1 to 0.3)	0.6 (±2.3)	1.6 (±5.6)	−1.0 (−2 to 0)
	Nursing home admission‡	0	0	—	2 (2)	9 (7)	0.2 (0.04 to 0.96)
	120-d service use† (%)	71 (39)	70 (39)	1.0 (0.7 to 1.5)	86 (59)	89 (61)	0.9 (0.6 to 1.4)
	Subsequent ED use	55 (31)	58 (32)	0.9 (0.6 to 1.4)	66 (45)	70 (48)	0.9 (0.6 to 1.4)
	Hospitalization	37 (21)	29 (16)	1.3 (0.8 to 2.3)	54 (37)	58 (40)	0.9 (0.5 to 1.4)
	Total hospital days (±SD)	1.0 (±5.6)	0.6 (±2.8)	0.4 (−0.5 to 1.3)	1.8 (±4.4)	3.0 (±8.6)	−1.2 (−2.8 to −0.4)
	Nursing home admission‡	2 (1.3)	1 (0.7)	2.0 (0.3 to 22.4)	3 (3)	11 (10)	0.3 (0.07 to 0.94)
	Costs, $
	Subsequent ED costs at 30 d§	14 (±43)	10 (±40)	5 (4 to 13)	25 (±66)	17 (±49)	9 (5 to 22)
	Hospital costs at 30 d	256 (±1,581)	150 (1,098)	106 (−177 to 389)	804 (±3,018)	1,252 (±3,168)	−448 (−1,162 to 266)
	Subsequent ED costs at 120 d§	37 (±88)	29 (±69)	8 (−9 to 25)	63 (±133)	50 (±92)	13 (−13 to 39)
	Hospital costs at 120 d	1,086 (±4,309)	721 (±2,746)	365 (−386 to 1,115)	2,217 (±5,312)	3,276 (±9,577)	−1,060 (−2,845 to 726)
	*χ2 Tests or Fisher exact tests are used as appropriate for categoric variables. Wilcoxon rank sum tests are used for continuous variables. †Service use refers to subsequent ED use, hospitalization, or nursing home admission. ‡All nursing home results are based on patients with data present. §ED costs are based on patients not admitted to the hospital. For patients entering the ED and then admitted to the hospital, costs are rolled into hospital costs.

The mean SF-36 Physical and Mental Component subscale scores were compared between the 2 groups at 30 and 120 days, respectively (Table 5).

Table 5. SF-36 scores at 30 and 120 days by study group.

	Variable	Intervention Group, Mean (95% CI)	Usual Care Group, Mean (95% CI)	Difference, Mean (95% CI)
	SF-36 Physical Component
	At 30 d*	37.0 (35.8 to 38.3)	37.6 (36.3 to 39.0)	−0.6 (−2.5 to 1.2)
	Change from baseline	−0.5 (−1.5 to 0.6)	0.2 (−0.9 to 1.3)	−0.7 (−2.2 to 0.8)
	At 120 d†	38.8 (36.9 to 39.8)	38.2 (36.7 to 39.7)	0.2 (−1.9 to 2.3)
	Change from baseline	0.7 (−0.3 to 1.8)	0.4 (−0.8 to 1.5)	0.4 (−1.2 to 1.9)
	SF-36 Mental Component
	At 30 d*	50.0 (48.5 to 51.5)	51.3 (49.9 to 52.7)	−1.3 (−3.3 to 0.7)
	Change from baseline	−1.5 (−2.7 to −0.3)	−0.9 (−2.0 to 0.2)	−0.6 (−2.3 to 1.1)
	At 120 d†	52.0 (50.6 to 53.4)	51.4 (50.1 to 52.7)	0.6 (−1.3 to 2.5)
	Change from baseline	0.1 (−1.2 to 1.4)	−0.7 (−1.8 to 0.5)	0.7 (−1.0 to 2.5)
	*Intervention group (n=286); usual care group (n=279). †Intervention group (n=269); usual care group (n=267).

At the 30-day follow-up interview, participants were asked to rate their level of satisfaction on 6 items regarding the ED nursing care and amount of information received during their index ED visit. The mean score differed between the 2 groups on the following item: “The information nurses or social workers gave about agencies or organizations that could help with your needs after you left the emergency room.” Participants in the intervention group were more satisfied with this item compared with participants in the usual care group (3.41 versus 3.03; mean difference 0.37; 95% CI 0.13 to 0.62). The groups did not differ on any of the other satisfaction items related to nursing care.

Discussion 

The aging of the US population and the increase in chronic illnesses dictates the need for continuity of care that optimizes health care delivery processes and patient outcomes.34 This randomized clinical trial tested the effectiveness of advanced practice nurse comprehensive assessment of older ED patients with appropriate referrals to medical, social, and health care community agencies on discharge. The intervention resulted in a significant increase in follow-up referrals but proved ineffective on overall service use. However, it was effective in terms of fewer nursing home admissions and greater satisfaction with the process of ED discharge care. Health care costs and use and health-related quality of life were not affected by the intervention. Subgroup analyses of elders considered to be at high risk revealed fewer nursing home admissions and shorter hospital stays among the intervention group. A separate analysis of the ability of our 6-item risk factor ED nursing triage tool to categorize older ED patients as high or low risk for adverse outcomes and its utility in predicting health care use is underway.

The overall lack of effect of our intervention might be related to several issues, including modifying the transitional model to the ED setting, the selected outcome and the nature of the illnesses and needs of this population, and the study sites and participants. Naylor et al's28 transitional model of care for hospitalized elders used the advanced practice nurse for comprehensive assessment and discharge planning combined with home follow-up visits in the first few weeks after hospital discharge. The only other controlled trial of a registered nurse comprehensive assessment and referral program to reduce subsequent ED visits among older ED patients of which we are aware also had negative results.17 In that study, referrals were suggested to the patient, family caregiver, or both, and it was at their discretion to make contact and arrange follow-up with the community agency. Our intervention was structured to address this issue by initiating the process in the ED and making direct referrals. Although referrals were significantly higher, the lack of direct advanced practice nurse follow-up might have weakened the intervention effect in our study. Because of the high volume and turnover of older patients in the ED, ongoing advanced practice nurse management was not feasible. Rather, we opted to use existing community agencies and primary care providers for the ongoing management. This approach was deemed reasonable because, although some have demonstrated the effectiveness of a comprehensive assessment combined with limited follow-up or direct management of older patients,35, 36 others have not.21, 37 Finally, little is known regarding the type or extent of services subsequently provided to the older ED patient on return to the community. The negative findings might be potentially attributable in part to the variation among providers in their ability to effect significant changes among older adults who used the ED.

The nature of acute illnesses on a background of multiple chronic illnesses among older adults might also negate the effectiveness of ED-based interventions aimed specifically at reducing subsequent ED use by older adults. As age increases, so do rates of ambulance arrivals, emergency (versus nonurgent) visits, adverse drug reactions, ED medication use, and hospital admission from the ED.2 In a recent cross-sectional study of 9,784 community-residing adults older than 65 years, Shah et al3 found that variables representing need characteristics (eg, state of health, ability to perform activities of daily living) predicted ED use, whereas predisposing characteristics (eg, age, sex, education) and enabling characteristics (eg, income level, insurance) did not. Rosenblatt et al4 also noted a strong relationship between increasing severity of illness and ED use by older Medicare patients. Older adults' baseline health and functional status and their effect on ED use might represent a significant reason why our intervention had little effect on subsequent ED use or overall hospital or nursing home admission. An appreciable number of our participants had significant baseline health problems and functional dependence, as evidenced by the numbers who required assistance in activities of daily living and instrumental activities of daily living.

In a series of studies, McCusker et al38, 39 examined ED revisits and functional outcomes in older ED patients. They found that older adults returning to the ED within 30 days were more likely to be depressed, to take 3 or more prescription medications, to drink alcohol daily, to indicate a lack of social support, or to be hospitalized within the previous 6 months.38 If depression, polypharmacy, and alcohol use are verified by others to be important predictors of ED use among older adults, then targeted interventions aimed at these factors, regardless of settings, might reduce subsequent ED use. In a follow-up randomized trial,39 they tested the effectiveness of a screening tool, standardized nurse assessment of at-risk elders, referral to appropriate medical and community services, and follow-up by the intervention nurse for up to 2 months in 38% of the cases. Their intervention reduced the rate of functional decline but was ineffective in terms of patient and caregiver satisfaction, patient depression, and caregiver health. Whether the intervention also reduced subsequent health care use was not reported.

Additional threats to internal validity included sample size, self-report of outcomes, proxy respondents, drop outs, and site selection. We did not reach our recruitment goal of 800 participants. Given the similarity of outcomes in the intervention and usual care groups, it is unlikely that statistical significance would have been achieved with the projected sample size. Ascertainment of outcomes might have been affected by patient recall. Medical records at the 2 sites were used to supplement self-report for ED and hospital use, but this was not feasible for nursing home use. Importantly, self-reports of ED use, hospitalization, and nursing home use have been shown to be reliable40 and, in our study, had an 89% agreement rate. Similarly, the use of primary caregivers as a proxy respondent for confused or incapacitated individuals has been shown to be a reliable source of data on service use, physical function, mental function, and satisfaction with health care services.41, 42, 43

Although attrition rates were similar between the 2 groups, drop outs might have affected the nursing home outcome given the small number of nursing home admissions. Sensitivity analyses were conducted because of missing data; nursing home information was not known for 82 participants at 30-day follow-up and for 115 participants at 120-day follow-up. The nursing home results are sensitive to the missing data. One lost-to-follow-up intervention patient changing to nursing home use at 30 days would change the results so that the OR, which currently goes from 0.05 to 0.99, would include 1, the OR of no treatment effect. Therefore, this result should be viewed with caution. The overall service use outcomes at 30 and 120 days, the only other primary outcome affected by missing telephone surveys, is not sensitive to the missing data. First, some participants not followed up at either 30 or 120 days have known positive outcomes for ED use or hospitalization (known through the hospital records). Only 58 participants at 30 days and 57 participants at 120 days lack follow-up for the overall service use outcome because hospital records indicate no ED visit or hospitalization and we lack the telephone surveys for nursing home admission. For service use to reflect a difference in groups, more than 57% of the 28 intervention participants not tracked would have to be in nursing homes versus 0 of the 30 usual care participants. Clearly, this is unlikely, given that only 2% of all tracked participants were admitted to nursing homes within 30 days. An even greater difference would be necessary to see a difference at 120 days. Therefore, we might be assured that the overall service use outcomes are not sensitive to the unknown nursing home use at 30 and 120 days.

The sites for this study might limit not only the generalizability but also might have influenced the outcomes. Both sites serve primarily indigent and minority populations. A recent Centers of Disease Control and Prevention report showed that higher rates of ED use by older adults were found among black and indigent populations.2 Lack of a regular physician-patient relationship4 and lack of supplemental insurance are also associated with higher ED use among older adults.2, 44

Additional threats to external validity include the locale, participant selection, and individual advanced practice nurses. The study took place in a county known for its richness and diversity of senior care services. The availability of existing social and health care agencies might be limited in other locales, limiting the feasibility of this particular model. The study participants were selected on the basis of their ability to respond to telephone interviews and residence within a certain geographic area of the ED sites. These methods of recruitment might have influenced the findings because eligibility for assessment and referral was based on selection criteria rather than requests from emergency physicians and nurses. This decision was made to minimize the potential for referral bias but might have weakened the effect of the intervention. Selection of most participants also occurred on the basis of their visits to the ED between the hours of 8 AM and 5 PM during weekdays. However, because the intervention was geared for those returning home, these are the hours when the majority of elderly ED patients who return home come to the ED. Those who come later in the evening and on weekends are more likely to be admitted to the hospital. Importantly, the intervention is likely to be relevant to only a portion of the elderly in the ED and hence requires a targeted approach. Finally, the 2 advanced practice nurses involved in the project were both graduates from a renowned school of nursing, each having more than 12 years of experience as geriatric advanced practice nurses in academic settings. The extent to which their skills and expertise are comparable with those of other advanced practice nurses is unknown.

In summary, this randomized controlled trial of an ED-based transitional model of care for assessment and linkage of community resources for community-residing older ED patients proved ineffective in overall service use of subsequent ED visits, hospitalizations, or nursing home admissions within 30 and 120 days after the index ED visit. It appeared effective, however, in reducing nursing home admissions at 30 days after the index ED visit. Importantly, subgroup analyses of the high- and low-risk groups demonstrated the effectiveness of the intervention for the high-risk patients by lowering the likelihood of nursing home use at 30 and 120 days and shortening hospital stays compared with those in the control group. Medicare expenditures for skilled nursing care after hospitalization quadrupled from $2.8 billion in 1989 to $11.3 billion in 1996 and are expected to continue to increase.45 Thus, the subsequent reduction in burden and long-term care costs is not an insignificant result and should not be discounted. The growing population of older adults and their increasing use of the ED compels ongoing examination into programs or models of care that would provide better assessment and management of this population's complex medical and social health care needs.