INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Intensive care units (ICUs) in rural hospitals can be characterized as low-volume units. Low patient volumes may be related to a number of factors including size and characteristics of the service area population, inter-hospital transfer practices, and poor quality (1). Low-volume units may generate a service demand level insufficient to recruit and retain intensive care specialists, resulting in a staff lacking specialized training and extensive experience in caring for the wide range of the most ill patients. One result is the potential for less than optimal diagnostic and therapeutic patient care practices. Community hospital ICUs with low patient volumes may have low rates of necessary diagnostic tests and therapeutic interventions (2).

From the viewpoint of the survival of the rural hospital, it may be essential to retain and strengthen ICU services. Such services provide a potentially life-saving resource in close proximity to rural populations. Evidence indicates that mortality among acutely ill patients decreased significantly with the development of designated critical care areas (3, 4). Indeed, evidence suggests that many patients treated in critical care areas return to pre-admission living arrangements (5) or functional ability levels (6, 7). Furthermore, patients treated in ICUs frequently report moderate to high levels of quality of life post-treatment (7). Maintaining quality of care in the rural environment despite potential resource limitations thus remains an important goal.

A particularly demanding and resource intense form of ICU care is mechanical ventilation. Mechanical ventilation is used in about 10% of ICU patients, but consumes 50% or more of ICU patient days and ICU resources (10). The reasons for the resource intensity are the heavy technology and professional staff requirements, and the prolonged stays often required for these patients. Guidelines for the care of mechanically ventilated patients are extensive and demanding, including requirements regarding personnel, monitoring and equipment, support services, ventilator management, and general patient management (11).

Research on the characteristics of ICUs that lead to better outcomes has produced conflicting findings. Two multisite studies concluded that lower ICU mortality rates were associated with degree of interaction and coordination of the ICU staff, but were not associated with administrative structural variables (12, 13). One of these studies also found that a patient-centered culture, and strong physician and nurse leadership were also associated with lower mortality (13). It has been suggested that ICUs that possess advanced technology, practice sound management principles, and focus on a narrow range of medical conditions exhibit superior outcomes (14). Lower mortality rates have been associated with better process performance (12), and with full-time staffing by medical and surgical resident teams (15). Cohen and coworkers (16) found in a study of a large teaching hospital that reductions in duration of ventilation and in frequency of some invasive procedures was accomplished by placing an interdisciplinary team in the ICU to supervise ventilatory management. However, the study by Knaus and coworkers (12) found that a full-time intensive care physician team was not essential if nursing-physician coordination was strong. Differences in admission practices, technology, and local practice style may also contribute to differences in resource use and patient outcomes among ICUs (17).

Thus, rural ICUs with limited resources, while providing essential services, may be at risk for providing suboptimal care because of limitations in technology and in staff knowledge and skills stemming from low patient care volume. However, prior evidence (12, 13, 16) suggests that it may be possible to improve care by focusing on processes and coordination, without necessarily changing structural features, technology, or volume. Therefore, we developed an educational outreach program tailored to institution-specific capacity and practices, and designed to improve quality of rural ICU care. We examined the effects of this program through a randomized trial, testing the hypothesis that the outreach program would improve quality of care processes, reduce resource use (i.e., length of stay and days of ventilator use), and improve patient outcomes (i.e., discharge disposition, nosocomial events).

The outreach program is modeled after a neonatal outreach approach in place in Iowa since 1973 (18). Core features of the program include the use of a university-based team of specialists with strong clinical backgrounds sensitive to the practical problems of providing care in a rural setting, and visits by the team to the rural setting where records are reviewed and individualized feedback provided to local staff and doctors. Record audits with face-to-face provider feedback were used because of past studies demonstrating this approach as an effective means of changing physician behavior (19).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Design

Twenty rural acute care hospitals in Iowa were recruited into the study through contacts between one of the authors (J.F.) and the hospital administrators and/or medical directors. The hospitals included seven Health Care Financing Administration (HCFA) designated rural referral hospitals, which are larger than other rural hospitals with more specialized staff and technologies. Twelve of the 20 hospitals were assigned randomly to the intervention condition and eight to the control condition, based on statistical power analyses from pilot studies. Four of the 12 intervention hospitals were rural referral facilities, as were three of the eight control hospitals. Data on patient care processes, outcomes, and resource use were collected at all 20 hospitals prior to and after the educational program. The study was approved by the university's institutional review board.

Data Collection

A nurse was trained in the data collection protocol. During training the nurse and the members of the outreach team independently reviewed a sample of ICU charts from the academic medical center and completed the data collection protocol. The nurse and team compared and discussed their reviews, and resolved differences. During the rural hospital visits a sample of records was reviewed independently by both the nurse and the team; kappa interrater reliability for these records was greater than 0.90.

The data collection protocol was derived employing objective indicators established by the Task Force on Guidelines, Society for Critical Care Medicine (11). Using the indicators as a guide, the university specialist team developed more specific standards based on current practice. The protocol reflects basic processes of ICU care that should be delivered regardless of technological sophistication or patient mix. The protocol consists of 44 process indicators in seven major categories: laboratory work (broken down into initial lab work, daily lab work, and lab work at Days 5 to 7 of the ICU stay), nursing care, ulcer protection, dietary management, immobility management, ventilator management, and ventilator weaning. Operational definitions were established for all indicators and listed as footnotes on the protocol. Indicators within each category are described in Table 1.

The data collection protocol also included collection of patient gender and age, diagnoses, admission and discharge dates for the ICU and the hospital, type of insurance, Acute Physiology and Chronic Health Evaluation II (APACHE II) scores on ICU admission, days on the ventilator, resuscitation status (do not resuscitate [DNR]) and days, admission source (emergency room, floor, nursing home, other hospital), number of high-risk conditions, and discharge status (died, home, nursing home, tertiary care facility, or other facility). High-risk conditions that result in difficult ventilation were developed through consensus of the outreach team, and included acute respiratory distress syndrome (ARDS), status asthmaticus, neurologic catastrophe, multiple organ dysfunction syndrome (MODS), sepsis syndrome with disseminated intravascular coagulopathy (DIC), peak pressure greater than 50 with positive end-expiratory pressure (PEEP) greater than 15, complex chest trauma, failure to wean, and complex overdose.

We also measured the occurrence of 18 nosocomial events taking place after 24 h in the ICU (categorized into eight groups: pulmonary, gastrointestinal, ventilation and monitoring, renal, cardiovascular, infectious, nutritional, and other). The selection of these events was based on Pingleton's "State of the Art" review on complications of acute respiratory failure (22). There were two exceptions to the 24-h rule: infectious events were those occurring at least 48 h after ICU admission using Centers for Disease Control (CDC) criteria, and tracheal intubation complications were recorded prior to 24 h if intubation was performed in the ICU. Carefully developed criteria were used to assess nosocomial events. For example, in order for pneumonia to be confirmed, a documented diagnosis by the primary contact physician was needed, or the presence of: (1) positive chest X-ray or computed tomography (CT), (2) or rales or rhonchi, plus (3) three of the following: cough, purulent sputum, temperature greater than or equal to 38° C, and potential pathogenic bacteria isolated by culture from sputum or trachea.

The nurse visited each hospital and reviewed medical records of the most recently treated mechanically ventilated ICU patients. All hospitals were visited initially in 1993 or 1994. Patients were selected by International Classification of Diseases, Ninth Revision (ICD-9) codes that reflect use of mechanical ventilation (96.72ventilated greater than 96 h, 96.71ventilated less than 96 h, and 96.70period of ventilation unspecified). We focused on patients ventilated more than 96 h in order to have a sufficient period to collect data on patient management techniques and practice patterns, and in order to obtain a more homogenous group of patients across facilities and time. However, some of the smaller hospitals had few patients with the 96.72 ICD-9 code, and in these cases we selected all patients with ventilation codes; 91% of patients at the rural referral hospitals were ventilated for more than 96 h, versus 35% at the smaller rural hospitals. Rural referral status is thus an important potential covariate in examination of results.

Prior to the first visit, each hospital was sent a survey to collect demographic information on the hospital, ICU, and staffing characteristics, including information on the number of hospital and ICU beds, staffing in the areas of nursing, pharmacy, nutrition, and respiratory therapy, the number and specialty of physicians, and the number of ICU admissions and daily census. Information was also collected on attendance at the team visit meetings, and on any major staffing changes between the first and second data collection visit.

Outreach Program

The outreach team consisted of a university-based pulmonologist, a nurse (not the same person as the data collection nurse), respiratory therapist, dietitian, and pharmacist. Each was experienced in ICU care of mechanically ventilated patients. The team was assembled and led by the pulmonologist, and had worked together on several outreach visits to other rural hospitals before the study began.

For the 12 intervention hospitals, the visit by the data collection nurse was followed within a few days by a visit from the team. During the interim the data collection nurse and team met to discuss the nurse's findings in order to prepare the team for specific areas likely to be a focus of teaching at each facility. During the team's visit, the ICU was toured, and then the team reviewed a sample of the same medical charts as had the nurse, using the same data collection protocol.

Each member of the team then met with his or her peers (the physician met with physicians, the nurse with nurses, etc.). During these meetings the team member reviewed the findings, and offered concrete, practical suggestions for improvement. Team members always included comments on things that the rural hospital was already doing well, which helped to establish trust and encourage willingness to hear suggestions for improvement. Suggestions for improvement were made objectively and confidentially, in nonthreatening ways, never singling out individual patients or providers.

Within a week of the team's visit, a written report was sent to the hospital administrator, medical director, and directors of nursing, pharmacy, respiratory therapy, and nutrition services. The report summarized the findings, and presented the quantitative results, including the percentage of processes successfully done, number of patients treated and their length of stay and discharge status, and occurrence of nosocomial events. A graph was included in the report that illustrated percentage of processes of care successfully complied within each category, and compared the hospital's results with aggregated results of other participant hospitals. The report also included reprints of pertinent authoritative clinical articles that supported recommendations to physicians and hospital staff.

Finally, rural hospital staff and physicians had access to three other postvisit resources: a quarterly newsletter prepared by the data collection nurse and intervention team summarizing developments in critical care, invitations to relevant university-based seminars, and a telephone consultation service. Each member of the team allotted approximately 2 h per wk to manage phone calls from their rural hospital peers to consult on specific patients, equipment, or protocol questions.

One year after the baseline data collection, each of the 20 hospitals was revisited by the data collection nurse. Medical records of mechanically ventilated ICU patients treated in the interim between visits were reviewed using the same protocol.

Analysis

Items on the protocol included some single yes/no processes (e.g., was magnesium included in the initial laboratory tests) and some processes that may be done numerous times during a stay (e.g., were arterial blood gas levels obtained within 60 min after major changes in the ventilator settings). In calculating the proportion of each process completed successfully, denominators were sometimes events (e.g., major changes in ventilator settings), and sometimes were ICU days (e.g., recording daily weights). Then the average percent compliance was calculated for each patient, within each category (lab work, nursing, ulcer protection, dietary management, immobility management, ventilator management, and ventilator weaning). Nosocomial event rates were calculated as the sum of the events divided by the ICU length of stay, times 100. These rates were calculated for nosocomial events in each of the eight category areas. Patient level data were aggregated to means to the hospital level.

Tests of the outreach model were based on hospital-level means. The hospital, not the patient, is the proper unit of analysis because the intervention was directed at the hospital level, and because the particular patients treated were not the same in the pre- versus postintervention period.

Before testing effects of the intervention, and despite the fact that the study was a randomized trial, we examined a range of possible covariates to determine if there were significant differences between treatment groups over time. Tested covariates included patient age, patient gender, hospital rural referral status, number of patients treated per hospital (patient volume), APACHE II admission score, presence of high-risk conditions, diagnostic groups (circulatory, respiratory, digestive, and injury), Medicare patient (yes/no), admitted via the floor (yes/no), admitted via the emergency room (yes/no), staffing changes, and percent DNR patients.

Dependent variables were the measures of process, resource use, and outcome. We examined the distribution of all variables and concluded that linear analyses without transformations were appropriate. Finally, we calculated the pre- to postdifference in covariates, and on the process, resource use, and outcome measures, and conducted t tests of the group differences.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Medical record data were collected from 224 patients in the preintervention period (135 patients in intervention hospitals and 89 patients in control hospitals). In the postintervention period data were collected from 170 patients (104 in the intervention hospitals and 66 in the control hospitals). Results from the previsit demographic survey are summarized in Table 2. They reveal a wide range of institutional resource capability. Hospital beds ranged from 29 to 320, for example, and ICU beds from 3 to 16. The total number of monthly ICU admissions ranged from 5 to 112, with an average of about 43. 

Comparisons of patients in the intervention and control groups at pre- and postintervention indicated that the two groups were highly similar (Table 3). There were no significant pre- to post hospital-level differences in any of the covariates. Therefore, tests of program effects were conducted as simple pre-post t tests without covariates.

The tests of intervention effects on patient care processes, outcomes, and resource use are summarized in Table 4. Results indicate significant (p < 0.05) program effects in the following processes: laboratory work, nursing, ventilator management, and total processes. Marginal effects (p < 0.10) were found for ulcer management and ventilator weaning. Compliance with ventilatory management recommendations, for example, improved from 51% in the preintervention period to 67% in the postintervention period, compared with 52% and 49%, respectively, in the control group.

Regarding measures of resource use, there was a marginally significant decrease in days on the ventilator (p < 0.09) associated with the intervention. There was an average hospital level decrease of 2.1 ventilator days in the intervention group, compared with a decrease of 0.1 d in the control group. Both total length of stay and ICU length of stay fell markedly in the intervention group, an average of 3.2 and 2.1 d, respectively, while the control group fell only 0.6 and 0.3 d, respectively; however, the effects were not statistically significant, due to the degree of variability in these measures across hospitals.

Regarding outcomes, there were no significant effects of the program on mortality rates, or on percentage of patients discharged home. Nor were there statistically significant effects for nosocomial events. Visual inspection of Table 4 results shows that observed rates in some event categories appeared to decline more in the treatment versus control groups (pulmonary, gastrointestinal). The small sample size and the relatively small number of events taking place across hospitals contributes to the difficulty in detecting program effects.

Qualitative findings indicate that two hospitals in the study lost their pulmonologist between the first and second data collection visits. One hospital gained a pulmonologist during this period, and one gained an internal medicine physician. These changes were equally split between intervention and control hospitals, and upon investigation seemed to have no effect on the results. Attendance figures at the educational sessions were high. In the 12 intervention hospitals, the programs were delivered to a total of 122 physicians, 111 nurses, 32 respiratory therapists, 15 pharmacists, 23 dietitians, and 11 other staff persons. The telephone consultation service was used 70 times, including 34 calls to the nurse, 28 to the dietitian, five to the physician, two to the pharmacist, and one to the respiratory therapist.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Results show that the rural ICU outreach program improved processes of care, and may have made care more efficient (i.e., reduced days of ventilator use). Specific types of recommendations made by the ICU team had the potential to decrease ventilator days. Our recommendations often addressed the same factors identified by Cohen and coworkers (16), including weaning at inappropriate speeds, weaning too late in the day, leaving patients on ventilators unnessarily overnight or over weekends, improving consistency of physician-nurse communications, and improving consistency of performing and recording weaning variables. Specific recommendations included appropriate selection of endotracheal tube size, increased use of appropriately set PEEP, and close monitoring of patient tolerance during weaning. Failure to recognize and respond to signs of respiratory muscle fatigue can prolong the weaning process significantly. The need for this type of fine tuning in management techniques would not have been recognized by the teaching team without first reviewing the medical record.

Additional improvements may have resulted from the rural provider's better understanding and awareness of practice guidelines developed by subspecialists. The recommendations and guidelines available in subspecialty journals are not always accessed at the rural level. Recommendation and sharing of algorithms, protocols, and standing admission orders created the momentum for change in many hospitals.

Despite the ability to improve most processes, the ICU team was unable to significantly improve nosocomial events, mortality, and length of stay outcomes. This is partly a statistical power problem. It remains unclear whether improving specific processes can change specific outcomes in the rural ICU setting.

That the program had no discernible effect on mortality or discharge home may also reflect the presence of patients with multiple organ diseases that cannot be treated successfully. Mortality is a rather insensitive indicator of quality, as patients near death may not survive despite exemplary care. Prior research has also demonstrated that the link between process quality and mortality is not particularly strong (23, 24).

This outreach educational program is an innovative approach through which the quality of low-volume, highly specialized patient care service can be improved in the rural ICU setting. Feedback from chart reviews of actual patient records, based on compliance with recommended guidelines, proved a powerful incentive for changing physician and staff behavior. This type of feedback was personalized to all participants because it impacted on real situations involving their patients. The feedback was delivered to groups of similar disciplines to allow for appropriate group dynamics. This also provided a comfortable group for discussion of key points. The small group sessions provided a forum to present a balanced discussion of quality and suboptimal patient care.

There was wide acceptance of the program, and the nonthreatening and personalized manner of the approach won the confidence of care providers and administrators. This approach, creating an atmosphere of cooperation and mutual respect between the academic center and the rural hospitals, demonstrates that successful collaboration has the potential to improve performance.