Time line and sample development. PCPs indicates primary care physicians; SCC, Senior Care Connections.
Sommers LS, Marton KI, Barbaccia JC, Randolph J. Physician, Nurse, and Social Worker Collaboration in Primary Care for Chronically Ill Seniors. Arch Intern Med. 2000;160(12):1825-1833. doi:10.1001/archinte.160.12.1825
To examine the impact of an interdisciplinary, collaborative practice intervention involving a primary care physician, a nurse, and a social worker for community-dwelling seniors with chronic illnesses.
A concurrent, controlled cohort study of 543 patients in 18 private office practices of primary care physicians was conducted. The intervention group received care from their primary care physician working with a registered nurse and a social worker, while the control group received care as usual from their primary care physician. The outcome measures included changes in number of hospital admissions, readmissions, office visits, emergency department visits, skilled nursing facility admissions, home care visits, and changes in patient self-rated physical, emotional, and social functioning.
From 1992 (baseline year) to 1993, the two groups did not differ in service use or in self-reported health status. From 1993 to 1994, the hospitalization rate of the control group increased from 0.34 to 0.52, while the rate in the intervention group stayed at baseline (P=.03). The proportion of intervention patients with readmissions decreased from 6% to 4%, while the rate in the control group increased from 4% to 9% (P=.03). In the intervention group, mean office visits to all physicians fell by 1.5 visits compared with a 0.5-visit increase for the control group (P=.003). The patients in the intervention group reported an increase in social activities compared with the control group's decrease (P=.04). With fewer hospital admissions, average per-patient savings for 1994 were estimated at $90, inclusive of the intervention's cost but exclusive of savings from fewer office visits.
This model of primary care collaborative practice shows potential for reducing utilization and maintaining health status for seniors with chronic illnesses. Future work should explore the specific benefit accruing from physician involvement in the collaborative practice team.
ALTHOUGH MEDICARE funds a vast array of health services for elderly and disabled persons, the coordinated delivery of these services remains a challenge. Moreover, it has lacked preventive services explicitly geared to reducing hospitalizations and premature admissions to nursing homes.1- 3 To correct such deficits, the social health maintenance organizations (HMOs) of the 1980s linked the HMO concept with case management and long-term care services for seniors. Unfortunately, they failed to achieve cost savings and improve care outcomes. The lack of strong physician involvement in treating high-risk patients and communicating regularly with case managers was seen as part of the problem.4,5
In the managed care environment of the mid-1990s, HMOs once again emerged as solutions for improving care and controlling Medicare costs.6,7 What should be the nature of the physician's role in caring for the elderly? Despite the belief that physicians' healing roles reside in the physician-patient relationship,8- 10 we have found no studies testing models that explicitly capitalize on seniors' physician-patient relationships, while simultaneously providing primary care and health promotion through formal collaboration with other health care professionals.
We demonstrated an 18-month intervention entitled Senior Care Connections (SCC) that asked primary care physicians (PCPs) in private practice to collaborate with specially trained nurses and social workers, as new practice associates, in the care of community-dwelling senior patients with chronic illnesses and functional deficits. We evaluated the impact of the SCC by assessing its overall acceptability and analyzing changes in the patients' use of services and in self-reported health status.
In mid-1992, 30 PCPs, 10 from each of 3 San Francisco Bay (California)–area counties, were invited to participate. They were preselected by local medical leaders as respected clinicians in their communities. Eighteen PCPs with sufficient patients to recruit for the study accepted. The 13 internists and 5 family physicians were randomized, by means of a random number table, to a control or intervention group. This resulted in 3 intervention and 3 control physicians per county or site. (At site C, 2 family physicians and 1 internist were randomized to intervention, while 2 internists and 1 family physician were randomized to control.) Each site and experimental group had 1 female PCP. We randomized physicians and not patients because of the difficulties anticipated in providing services to some patients and not others in the same office practice. Median physician age was 52 years, with 14 years in practice. None of the physicians were partners of each other. Two physicians per experimental group were solo practitioners, and the others were in practices ranging in size from 3 to 6 partners. The median number of patients per day was 13.5; 4.5 were aged 65 years or older.
Before being randomized, each physician met with the study coordinator and used preset criteria to select at least 35 study patients from a list of patients seen consecutively in their office during the past 2 months (see below). The objective was to define a group of patients who were community dwelling but had difficulties in living independently.11
One or more visits with primary care physician between June 15, 1992, and September 15, 1992
Age 65 years or older as of January 1, 1992
Spoke English well enough to communicate with the nurse and social worker
Independent in the following activities of daily living: walking (walking devices permitted), transferring, toileting, and feeding (patients were considered eligible if they needed assistance in bathing and dressing, but they could not require 24-hour attendant care)
Unable to carry out at least 1 instrumental activity of daily living: getting around outside the home, meal preparation and household chores, taking medications, use of the telephone, and money management
Not terminally ill
Not residing in a nursing home (a residential care facility for the elderly was allowable)
Not under therapy for metastatic disease, Alzheimer disease, or related dementias
Under treatment for at least 2 chronic conditions (stable or unstable)
If both chronic conditions were stable, having at least 1 health risk factor (sedentary lifestyle, hyperlipidemia, obesity, smoking, alcoholism, social isolation, depression, anxiety)
After physician randomization, in November 1992, 351 control patients and 383 intervention patients received a 20-page questionnaire with a cover letter from their physician. Patients were asked for demographic data and were queried about daily habits (eg, exercise, smoking, alcohol use), use of support services (eg, home-delivered meals), chronic conditions, and self-efficacy for health-related behaviors.12 Their physical functioning was assessed via the Health Activities Questionnaire,12 emotional functioning via the short form of the Geriatric Depression Scale,13 and perceived health status by means of the Medical Outcomes Study 36-Item Short Form Health Survey health status question.14 Checklists were used to assess nutritional habits,15 recent symptoms,16 and social activities,17 and a list of current medications was requested.
Eighty-seven percent of intervention patients (333) and 78% of control patients (274) returned the first questionnaire (P=.002; see Figure 1 for study time line and sample development). During the 6-month enrollment period beginning in January 1993, as intervention patients came to the offices for their scheduled appointments, the PCPs determined whether they still met study criteria and, if so, described the SCC and introduced the nurse and social worker. By April 1993, the PCPs found that 238 (71%) of intervention patients still met study criteria and agreed to participate. Twenty-two patients (7%) refused participation. The remaining 73 patients (22%) did not meet criteria for the following reasons: 20 (6%) had become "too frail" to take advantage of the SCC (eg, 24-hour care needed at home); 40 (12%) were stable clinically and functioning at a high level, invalidating their need for the intervention; and 13 (4%) had died, switched physicians, or moved out of the area within the preceding 6 months. These patients were excluded from further study participation, and no further data were obtained about them. On the basis of demographic and health status data from the first questionnaire, excluded patients differed from other intervention patients on only 1 variable: a higher likelihood of being married as opposed to single (60% vs 44%; P=.01).
To obtain an identifiable patient cohort of 30 patients within the practice, each intervention PCP extended participation to patients not originally sent the first questionnaire, but who were seen in the office during the enrollment period and met study criteria. Forty-two patients were added in this way (2 to 6 per PCP) and represented 15% of the 280 intervention patients. These patients did not differ from original intervention enrollees on baseline measures with the exception of having a 14% hospital readmission rate in 1992 compared with the original patients' rate of 3% (P=.01).
Control physicians did not re-review patients as they came in for office visits during the enrollment period, but notified the study coordinator about the 4% of patients who had died, moved out of the area or to nursing homes, or switched physicians during the preceding 6 months. No new patients were added to their lists of study patients, which averaged 26 patients per control physician (range, 21-37) and resulted in a total of 263 control patients. After the 1992 baseline year, the SCC began in January 1993 and continued through June 1994. Patient enrollment occurred during the first half of 1993; the last half of 1994 was a follow-up period with no SCC services.
This office-based intervention, described previously,18 demanded close collaboration among a PCP, a registered nurse with geriatrics training, and a master's-prepared clinical social worker experienced in working with seniors and their emotional health concerns. Each full-time nurse and half-time social worker divided his or her time among 3 intervention physicians whose individual practices were located in 1 San Francisco Bay–area county. A PCP, a nurse, and a social worker made up 1 office-based team; across the 3 counties, there were 9 teams. Before the SCC start date, the nurses and social workers participated in an intensive, 2-month immersion in their physicians' practices. Throughout the intervention, they met with trainers to learn team-building skills and strategies for coaching patients in chronic disease self-management.19
The SCC intervention focused on a set of defined activities for each intervention patient. First, the nurse or social worker visited the patient in the home, listened to health concerns, took vital signs and health histories, and completed a patient functional assessment and a home safety check. Second, using these data and the physician's prior knowledge of the patient, the team discussed the patient's health status and generated frailty and health risk scores. They drafted a risk reduction plan for discussion with the patient and family to set target objectives (eg, reduce arthritic pain) and plan treatment by means of chronic disease self-management strategies (eg, patient contracts to take 4 short walks daily). Third, the nurse and social worker monitored the patient's health status between office visits through contacts by telephone, home visit, small-group session, or office or hospital visit at least once every 6 weeks. During contacts, the nurse or social worker inquired about new problems, checked chronic disease status, coached patients in self-management skills, and promoted use of community-based services (eg, home-delivered meals). Finally, the physician, nurse, and social worker met at least monthly to review each patient's status and revise care plans.
The study had 2 categories of end points: utilization of medical services and patient-reported health status. To measure change in utilization, we assessed the following: (1) number of hospital admissions per patient; (2) number of patients with 1 or more hospital readmissions within 60 days; (3) mean number of office visits to all physicians, to study PCPs, to medical specialists, and to other non–primary care, non–medical specialty physicians (eg, surgeons, orthopedists, ophthalmologists, dermatologists, psychiatrists, and physiatrists); (4) number of patients with 1 or more visits to the emergency department; (5) proportion of patients with 1 or more home care visits, and (6) number of patients with 1 or more nursing home placements. We designed these measures by means of data acquired from the Health Care Financing Administration's (HCFA's) National Claims History Database and from equivalent administrative databases of the Aetna and the QualMed Medicare HMOs for calendar years 1992, 1993, and 1994. Data were available for the assessments described above from both HCFA and the Medicare HMOs with the exception of emergency department, home care, and skilled nursing facility (SNF) data from the Medicare HMOs. A claims validation process of 100% of hospital admissions (using hospital medical records) and a 10% random sample of physicians' office visit claims for each study year (using physician office records) resulted in overall congruence rates of 92% for hospital admissions and 75% for office visits.
Medical service utilization data for the 3 study years were available for 465 (86%) of the 543 patients in the total sample; 15% of control patients and 14% of intervention patients lacked these data for each study year (P=.69). These patients included (1) 18 federal or railroad workers whose claims were not processed by HCFA or Medicare HMOs, (2) 34 patients who died in 1993 or 1994, and (3) 26 patients for whom neither HCFA nor the HMOs provided data for 1993 and/or 1994, although data had been provided for 1992. Patients with and without 3-year data did not differ on any baseline utilization or health status measure. Of the 465 patients with 3-year data, 77% were standard Medicare patients; 23% were Medicare HMO patients, with equivalent proportions in each experimental group (P>.99). Traditional Medicare and HMO patients did not differ on any demographic or health status measures at baseline. The HMO patients did, however, have a higher hospitalization rate (45% vs 34%; P=.06), a larger proportion with 1 or more 60-day readmissions (9% vs 4%; P=.04), and more office visits (13.6 vs 12.5; P=.03).
Patient-reported health status data were obtained via 3 questionnaires mailed in fall 1992 (baseline), 1993, and 1994. All 3 questionnaires were returned by 384 patients: 70% of the control patients and 72% of the intervention patients (P=.78), with equivalent proportions of traditional Medicare patients (73% and 74%) and Medicare HMO patients (P=.82). The remaining 159 patients included 41% who did not return the second and/or third questionnaires despite numerous reminders (14% of control patients and 10% of intervention patients); 33% who died in 1993 or 1994 (10% and 9%, respectively); 16% who changed PCPs (5% and 4%, respectively); 6% who moved to nursing homes (0.8% and 3%, respectively); and 4% who moved out of the area (0.4% and 2%, respectively). After removal of the 39% who died or went to a nursing home (a significantly older group with more chronic conditions and poorer function at baseline), the remaining patients lacking 3 questionnaires did not differ from those with 3 on any baseline utilization or health status measure.
The nurses and social workers entered data about their patient contacts into laptop computers as close in time to the patient contact as possible.
Baseline differences between the intervention and control groups were analyzed by means of the 2-way χ2 test, Fisher exact test, Mann-Whitney tests for ordinal data, and independent group t tests for continuous data. The analysis of end points was 2-pronged. First, using a sample made up of all 465 patients having 3-year utilization data, we examined treatment group differences in amount of change observed in hospital admissions, office visits, emergency department visits, home care visits, and SNF admissions, between 1992 and 1993, and between 1993 and 1994, adjusting for 1992 baseline levels. Second, using a sample made up of 384 patients having 3-year questionnaire data, we examined treatment group differences in the amount of change observed between the same time periods for physical, emotional, and social functioning; nutritional habits; symptoms; use of medications; and overall health status, adjusting for 1992 baseline levels.
The generalized estimating equation method20,21 was used to assess the influence of the SCC on utilization and health status end points. It modeled outcomes in adjacent years (1992 and 1993, 1993 and 1994) from treatment group, year, PCP's site, treatment group by site, treatment group by year, physician within treatment group, patient age, sex, 1992 health status, and 1992 rate or mean for the targeted dependent variable. It was assumed that patients of one PCP had more in common with each other than with patients of another PCP; physician predictors were introduced to eliminate within-physician correlation between subjects. Marital status and support service use was also added to the model to adjust for baseline group differences. To analyze group change rates over time, the primary effect studied was the treatment group×year interaction.
Analyses of hospitalization and office visit counts used a Poisson data model with a log link function. Office visit counts were first log transformed, adding 1 to deal with zero-visit cases, since models on the untransformed counts failed to converge. Analyses of binary outcomes (eg, ≥1 emergency department visits) used a binomial data model with a logit link function. Analyses for continuous variables (eg, depression score) used a model for normally distributed data.
In the absence of a validated, global health status measure for seniors, we constructed a composite health status score to adjust for baseline health status. This score was derived by means of a principal-components analysis on the correlation matrix of individual scores obtained from the Health Activities Questionnaire, Geriatric Depression Scale, 3 checklists (social activities, symptoms, nutritional habits), a medication list, and the Medical Outcomes Study 36-Item Short Form Health Survey health status question.22 The resulting principal-components score was used to adjust for baseline functioning and health status in the subsequent analyses of differences in group change rates. This score at baseline had a mean value of +0.13 for nonhospitalized patients and −0.37 for hospitalized patients (P=.005; positive values indicate better status).
For 1994 only, using independent group t tests for continuous data, we looked at group differences in length of hospital stay, available for all admissions, and hospital admission payment amount, available for 67% of Medicare admissions. Office visit payment data were not available.
P values less than .05 are statistically significant in comparisons between groups; group differences with P values less than .10 are reported as trends. Odds ratios (ORs) for proportions with 95% confidence intervals are given for all differences except office visit rates; log transformations made these calculations infeasible.
Intervention and control group demographics were similar with the following exceptions: control patients tended to be younger, more likely to be married or not live alone, and less likely to use support services (Table 1). Intervention and control patients did not differ on any utilization or health status measures in 1992 (Table 2 and Table 3).
Throughout 1993 and 1994, the PCP office staffs reported patient deaths. Ten control patients (3.8%) and 12 intervention patients (4.3%) died (P=.83) in 1993, and 16 control patients (6.3%) and 14 intervention patients (5.2%) died (P=.71) in 1994. Vital status at the end of 1994 could not be obtained for 8 control patients and 6 intervention patients who had either moved or switched PCPs.
The nurse or social worker had at least 1 contact with 85% of all intervention patients by May 1993. Beginning with the first contact, the average patient experienced 14 months of the SCC (range, 10-18 months); 14 patients had less than 12 months, of whom 13 were patients of the same physician. During the 18-month SCC, all 280 intervention patients had at least 1 face-to-face contact (other than the initial assessment visit done in the home) with the nurse or social worker. Patients averaged 34 nurse or social worker contacts (range, 1-176). Seven percent of patients received less than 10 contacts, 47% received 10 to 29 contacts, and 46% received 30 to 176 contacts. On average, contacts lasted 22 minutes and occurred every 21 days. Nurses initiated 58% of contacts, 26% of contacts were initiated by social workers, and 16% were initiated by patients. Sixty-nine percent of contacts were by telephone, 17% by home visit, 4% by office visit, 4% by mail, 3% by hospital visit, and 3% in small-group sessions. In fall 1993 and fall 1994 (after the SCC had ended), intervention patients were asked via a 1-page, mailed questionnaire: "In general, how would you rate the usefulness of the SCC?" Response rates for each year were 81% and 88%, respectively, with mean scores of 3.9 and 4.0 on a 5-point scale, a 5 being "extremely useful." Four intervention patients (1.4%) dropped out, stating that they did not need the SCC; 1 control patient refused subsequent questionnaires.
During the 18 months, each physician–nurse–social worker team formally met 24 times on average (range, 16-35). Additionally, 9 educational sessions taught by geriatricians were well attended by participating clinicians; a half-day retreat was held in the ninth month. All 9 intervention physicians, 3 nurses, and 3 social workers requested that the SCC continue past the original 18 months; replication efforts of parts of it were initiated in 2 of the 3 counties and funded locally.
From 1992, the baseline year, to 1993, year 1 of the SCC, we found no significant differences at P<.05 between control and intervention groups in amount of change in measures of hospital admissions, readmissions within 60 days, office visits, emergency department visits, home care visits, or SNF admissions (Table 4).
From 1993 to 1994, year 2 of the SCC, we observed a stable hospital admission rate for intervention patients (0.38 to 0.36) compared with controls, whose rate increased from 0.34 to 0.52 (P=.03, OR, 0.63; 95% confidence interval, 0.41-0.96). In addition, intervention patients showed a 6% to 4% decrease in the percentage with 1 or more hospital readmissions within 60 days contrasted to the controls' increased proportion (4% to 9%; P=.03; OR, 0.26; 95% confidence interval, 0.08-0.84). A diagnosis-related group analysis of the control and intervention groups' hospitalizations in 1994 showed that 64% of control patients' admissions were assigned medical diagnosis-related groups, of which 22% were for 4 common chronic illness conditions: heart failure, chronic pulmonary disease, stroke, and nutritional abnormalities. In contrast, 56% of intervention patients' admissions were assigned medical diagnosis-related groups, of which 13% were for chronic illness conditions.
For total physician office visits, we observed a decrease in intervention patients' mean office visits of 1.5 visits (12.5 to 11.0) to any type of physician compared with a 0.5-visit increase for controls (P=.003; OR was not interpretable because analysis was done on the log of office visits). This difference in group change rates for total office visits came largely from intervention patients' fewer visits to non-PCPs, either medical subspecialists or physicians such as orthopedists or dermatologists (Table 5). Decreased visits to PCPs were evident in both groups, but group change rates were not significantly different at the P<.05 level (P=.50). No differences were evident in group change rates from 1993 to 1994 in proportions of patients with 1 or more emergency department visits, home care visits, or SNF admissions.
Differences between groups first became apparent during the final 6 months of the SCC (January to June 1994). At this time the control group's hospitalization rate increased to 0.30 in contrast to 0.14 for the same period a year earlier; this increase contrasts with the intervention group, for which the rate remained stable at 0.16 for both periods (P=.01). Similarly, the control group's mean office visits for this period increased to 7.1 compared with 6.4 a year earlier. In contrast, the intervention group's mean office visits fell to 6.4, down from 6.8 (P=.005). During the follow-up period (July to December 1994), group differences in hospitalization rate changes were no longer significant, but intervention patients continued to stay ahead of control patients in declining office visits (6.4 to 5.2 as compared with 6.3 to 6.0; P=.04).
In 1994, hospitalized control and intervention patients did not differ in length of hospital stay (6.0 and 6.3 days) or in cost of the average admission ($9138 and $9959). Since intervention patients had 26 fewer admissions during 1994, a total of $258,934 was saved. Costs of implementing SCC for 1994 totaled $118,950 including salaries and benefits of nurses and social workers, plus overhead and training costs. Subtracting implementation costs, net per-patient savings in 1994 were $90, not inclusive of savings from fewer physician visits.
No significant differences at P<.10 were observed in group change rates from 1992, the baseline year, to 1993, year 1 of the SCC, for the 7 measures of health status (Table 6). In contrast, from 1993 to 1994, we observed a higher mean number of social activities for intervention patients (8.6 to 8.8) compared with controls (8.9 to 8.6; P=.04; 95% confidence interval, 0.02-0.1.0). In addition, compared with controls, we observed trends for intervention patients to report fewer symptoms (17.7 to 17.2 vs 17.9 to 18.9; P=.08) and to have slightly improved overall health (3.2 to 3.2 vs 3.2 to 3.3; P=.08). No differences in group change rates were observed for physical, emotional, or nutritional status, or number of medications.
The SCC intervention explored the physician-patient relationship as a base for the evolution of a team-patient relationship. Once established, the team-patient relationship showed potential for effecting less use of acute care services and physician office visits while maintaining, if not improving, patient-perceived health status. In this age of managed care, interdisciplinary teams in geriatrics are receiving increasing attention. Despite the earlier failures of the social HMOs, several smaller trials have successfully tested the role of geriatric nurse practitioners and interdisciplinary teams in reducing seniors' hospital admissions and improving health status. Interventions tested include many of those packaged in the SCC: preventive home visits,23- 29 assessment and service coordination efforts,30- 32 nurse or multidisciplinary team assessments with home follow-ups,33- 42 and health promotion programs.43- 45
To date, however, none of the above interventions has highlighted the physician role or that of the physician-patient relationship. Yet active physician involvement in collaborative caregiving has been associated with improved patient outcomes in intensive care units,46,47 other inpatient settings,48 general clinic populations,49 and residential treatment centers.50 The level of physician interaction with nonphysician caregivers in the SCC intervention—introducing patients to nurses and social workers as clinician associates, staying in regular contact, having monthly meetings to assess patient status and revise care plans—has not been reported elsewhere.
The fact that differences between the control and intervention groups did not surface until the last 6 months of the intervention should not be surprising. Through personal interviews and patient satisfaction questionnaires, participants acknowledged that the first 12 months was largely spent developing a trusting relationship with the team. Clinicians and patients needed time to get to know one another and test communication modes (eg, voice mail, faxing); collaboration could not be engineered via algorithm to surface on demand.
As regards the SCC's possible mechanism of impact, preliminary investigation suggests a dose-response relationship between number of nurse and social worker contacts and both service utilization and patient health status. Using analysis of variance with a test for linear trend in the means, we saw that statistically significant, positive relationships emerged between contact number (none, low, medium, or high) and utilization as well as health status change rates (Table 7). For example, control patients with no contacts had the largest hospitalization rate increase (0.17), intervention patients with less than 21 contacts had a 0.07 increase, patients with 21 to 38 contacts had a 0.05 increase, and patients with more than 38 contacts had a hospitalization rate decrease of 0.18. The diagnosis-related group analysis of hospital admissions suggested that, compared with controls, intervention patients had a lower proportion of admissions for exacerbation of chronic illnesses. Were patients with more contacts seen as being at higher risk for hospital admission? Did these patients, in turn, have fewer admissions because of the increased attention by their PCPs, nurses, and social workers? Future research should look at the impact of nurse and social worker contacts on service use as mediated by patient chronic disease burden and functional status. In addition, key "relationship" variables need exploration, such as the strength of the original physician-patient relationship, the patient's satisfaction with nurse and social worker contacts, and the quality of the PCP's, nurse's, and social worker's collaborative relationship. Preliminary analyses by geographic site suggest that differences in the hospitalization rates between the control and intervention groups were the greatest (suggesting lower rates for intervention patients) in the county where the PCP, nurse, and social worker were the most satisfied with their working relationships.
Limitations in study design and implementation require a cautious interpretation of findings. Threats to generalizability include characteristics of the PCPs' practices (eg, well-established practices in an affluent region of California; willingness to open their practices to nurses and social workers) and attributes of the patients themselves (eg, white, middle-class, and not severely compromised functionally). In addition to the use of a small number of practices to randomize, threats to validity arose in 4 key arenas. First, since we did not obtain similar questionnaire return rates from patients in both experimental groups (78% and 87%, respectively), the experimental groups may not have been equivalent at baseline. (We suspected that intervention office staffs encouraged questionnaire return to expedite patient enrollment.) Second, only intervention PCPs, not control PCPs, reassessed patients for SCC appropriateness as they came into the offices 4 to 8 months after being initially selected. Through this process, intervention PCPs excluded 95 patients who had returned the initial questionnaire but, at the first 1993 office visit, no longer met criteria or refused SCC participation. The control physicians did not reassess their patients for continued study participation because we believed that a focused review would be a departure from usual practice. The 42 new patients whom intervention physicians added to round out their group size to 30 patients could indeed have created a cohort markedly different from the one that would have resulted from the original randomization. (These patients did have a higher baseline hospital readmission rate than the original patients.)
Although the outcomes of the 95 excluded patients were not known, the impact of the 42 new patients on the intervention group's change rates was investigated by removing them from the analyses. Similar differences in group change rates were still evident at P<.05 for hospital readmissions, office visits, and social activities, with similar trends for symptoms and self-rated health status. The difference in group change rates for hospitalizations, although still showing a trend, was no longer significant at the .05 level (0.33 to 0.50 for controls, 0.36 to 0.37 for intervention; P=.10).
Fourteen patients were unavailable for follow-up. We relied on PCPs' offices to report expirations and underestimated the number of patients who would move or switch physicians during a 2-year period and the effort required to track them. Finally, only 64% of the original patient sample had both 3 years of utilization data and 3 years of health status data. By using 2 samples—one for analyzing utilization change rates, the other for health status change rates—we may have overlooked systematic differences between samples. In addition, because of missing emergency department visit, home care visit, and SNF admission data for Medicare HMO patients, we could not conclude that differences in group change rates for these variables did not exist.
Future research should articulate the value of the PCP's involvement with the collaborative practice team and how the physician-patient relationship is changed by team participation. The results of this demonstration effort bode well for carrying out a randomized controlled trial that would implement a more rigorous design and explicitly test alternative models of physician involvement.
In summary, the SCC intervention successfully modeled interdisciplinary collaboration in the private office setting for chronically ill seniors. Furthermore, it suggested the potential for lowering hospitalization rates and reducing non-PCP office visits for these patients while maintaining function. Finally, it highlighted the practical dilemmas yet feasibility of practice-based research involving older persons. Campion51 said that the predicament of seniors is to find "the security of care" and place it literally and figuratively in their homes and communities. The SCC attempted to establish this security, using collaborative primary care practice as its platform. Our findings suggest that the platform could well be a solid one to build on.
Accepted for publication October 26, 1999.
This research was supported by a grant from the John A. Hartford Foundation, New York, NY (as part of their Generalist Physician Initiative Program), to the California Pacific Medical Center, San Francisco, with support from Alta Bates Medical Center, Berkeley, Calif, and Marin General Hospital, Corte Madera, Calif.
Presented at the Annual Meeting of the North American Primary Care Research Group (Outstanding Paper Award), Montreal, Quebec, November 8, 1998.
We are indebted to the following nurse, social worker, and physician participants: Stephen Becker, MD; Robert Belknap, MD; David Berman, MD; Suzanne Bourque, RN; Tom Bodenheimer, MD; Janet Bodle, MD; James Clever, MD; Catherine Clark-Sayles, MD; Anne Comer, MSW; Richard Dennes, MD; Kenneth Gjeltema, MD; Gretchen Kunitz, MD; Suzanne Leib, MFCC; Nori Mandell, RN; Guy Micco, MD; Sophie Mirviss, MD; Robert Pedrin, MD; Cynthia Point, MD; Joshua Rassen, MD; Victor Rosenoer, MD; Susie Scholefield, MSW; Mary Sears, MD; Rosemary Stevens, MD; Loren Stolle, MD; Michael Smith, MD; Gayle Taylor, RN; James Taylor; and Louis Wu, MD. Robert Escamilla; Doug Berman, MBA; and William Dapkus, MBA, provided technical support; Alan Bostrom, PhD, and David Heilbron, PhD, provided statistical consultation; and Brad Kirkman-Liff, DrPH; Hal Holman, MD; Kate Lorig, RN, DrPH; Ellen Netting, PhD; Donna Regenstreif, PhD; Mark Sager, MD; Anita Stewart, PhD; and Frank Williams, PhD, provided ongoing consultation.
Corresponding author: Lucia S. Sommers, DrPH, Associate Program Director, Internal Medicine Residency Program, St Mary's Medical Center, 450 Stanyan St, San Francisco, CA 94117-1079 (e-mail: email@example.com).