Printer Friendly

Identification of factors associated with hospital readamission and development of a predictive model.

Multiple hospital admissions, especially those to the chronically ill, represent a particular challenge to both the acute and long-term care sectors to identify effective methods of resource management. This study analyzes the multiple admission patterns associated with a cohort of 4,219 adult medical-surgical patients discharged alive from a community teaching hospital in Michigan. The sample was divided into two groups: 3,818 patients who survived and 392 who expired during the one-year follow-up period. For the surviving subsample, the characteristics found to be directly associated with the likelihood of readmission included increase age, advanced stage of disease, greater index-episode length of stay, discharge by an internist rather than a surgeon, Medicare as expected source of payment, decreased physician age, discharge to a community setting, and increased number of prior hospital episodes. For the subsample who died, only one explanatory variable was significantly associated with an increased likelihood of readmission--discharge to a community setting (with or without home care) rather than a nursing home. The article includes illustrations of the importance of decisions regarding posthospital, long-term care services on the likelihood of rehospitalization.

It has been recognized for some time that a small proportion of the population accounts for a disproportionate share of the consumption of health care resources (McCall and Wai 1983; Roos and Shapiro 1981). Approximately one out of every six or seven patients admitted to a hospital experiences a second admission within less than one year, and for Medicare beneficiaries the readmission rate may be as high as 22 to 27 percent within 60 days of discharge (Anderson and Steinberg 1984; Holloway, Thomas and Shapiro 1988).

Multiple admissions particularly concern providers, patients, and payers as possible indications of poor quality or inefficiency in providing patient care. To the extent that multiple admission are attributable to care of the chronically ill, they present a particular challenge to both the acute and long-term care sectors to identify effective methods of management.

The objectives of this analysis are to identify patient and health system characteristics associated with readmission, and to construct statistical models that can be used to estimate the probability of readmission for a given patient at some future point in time.

Literature Review

Numerous studies have been conducted to identify the relationship of various patient and health system characteristics to the likelihood of readmission (Graham and Livesley 1983; Anderson and Steinberg 1984; Victor and Vetter 1985; Roos et al. 1986b; Fethke, Smith, and Johnson 1986; Roos et al. 1986a; Holloway, Thomas, and Shapiro 1988; Holloway, Medendorp, and Bromberg 1990). Patient characteristics include demographic-social, clinical, and resource use characteristics. Health care provider and system characteristics include: physician specialty, age, experience, practice setting and caseload; hospital teaching status; and community supply of health resources.

Age has been found to be positively associated with the likelihood of readmission until old age is reached; at that time there appears to be no significant relationship between age and readmission (Holloway, Medendorp, and Bromberg 1990; Holloway, Thomas, and Shapiro 1988; Fethke, Smith, and Johnson 1986). Males have higher readmission rates than females in studies employing shorter time windows (e.g., less than three months), while studies with longer time windows (e.g., one year) found no significant gender difference (Graham and Livesley 1983; Anderson and Steinberg 1984; Fethke, Smith, and Johnson 1986).

Research has produced conflicting results regarding the effect of marital status on readmission, especially when various living arrangements and socioeconomic status are controlled for (Fethke, Smith, and Johnson 1986; Holloway, Medendorp, and Bromberg 1990). Several studies have examined the relationship between "living alone" and the likelihood of readmission and have reported insignificant findings (Graham and Livesley 1983; Victor and Vetter 1985; Fethke, Smith, and Johnson 1986; Holloway, Thomas, and Shapiro 1988). One study that included measures of income found no significant differences in readmission rates among individuals of various income levels (Fethke, Smith, and Johnson 1986), while another that sampled only from Medicare beneficiaries found that individuals having both Medicare and Medicaid coverage showed significantly higher rates of readmission than those with only Medicare coverage (Anderson and Steinberg 1984). Some evidence indicates that rural residence or increased distance from hospitals, or both, are associated with higher readmission rates (Anderson and Steinberg 1984; Holloway, Medendorp, and Bromberg 1990).

Variations in readmission rates across diagnostic groups and surgical procedures have been well documented, with patients having certain chronic conditions or undergoing certain surgical procedures exhibiting higher readmission rates (Riley and Lubitz 1986; Zook, Savickis, and Moore 1980; Victor and Vetter 1985; Holloway, Medendorp, and Bromberg 1990). Some studies have also documented an increased risk of readmission associated with the presence of multiple chronic conditions and the performance of multiple surgical procedures (Fethke, Smith, and Johnson 1986; Roos et al. 1986a; Holloway, Thomas, and Shapiro 1988; Holloway, Medendorp, and Bromberg 1990). Higher readmission rates have also been found to be associated with poorer health status and impairments of sight or hearing, motor disabilities, and incontinence (Holloway, Thomas, and Shapiro 1988; Anderson and Steinberg 1984; Victor and Vetter 1985).

Two studies have reported a direct relationship between prior hospitalization patterns and the probability of readmission (Fethke, Smith, and Johnson 1986; Roos et al. 1986a). No significant association has been found between the length of stay for the index episode and the likelihood of readmission (Victor and Vetter 1985; Holloway, Medendorp, and Bromberg 1990).

Very little information is available regarding the use of nonhospital services subsequent to hospital discharge and its effect on readmission. Holloway, Mendendorp, and Bromberg (1990) found no association between readmission and the use of nursing homes or other subacute institutional services subsequent to hospital discharge, but they did find that discharge from an intermediate care ward in a Department of Veterans Affairs medical center was highly predictive of early readmission.

In an analysis of the hospital utilization patterns of primary physicians, Roos and colleagues (1986b) found that, after adjusting for differences in case mix, increasing physician age was generally associated with (1) a lower percentage of patients in the practice who were hospitalized; (2) lower readmission rates; (3) an increased average length of stay; and (4) a lower mean number of hospital days per patient in the practice per year. Physicians with appointments in teaching hospitals averaged significantly fewer days of hospitalization per patient in their practices, and rural physicians exhibited higher hospital utilization rates in the form of higher percentages of patients hospitalized, but lower average lengths of stay. Roos et al. found no significant differences in the readmission rates between physicians in solo and in group practice, but they did find that increased bed supply relative to the population in an area was directly related to higher readmission rates (Roos et al. 1986b).

This study contributes to the current knowledge base regarding readmissions by validating some of the results of earlier studies, and by incorporating better measures of case mix, severity of illness, and long-term care utilization.


The data used in the analysis pertain to adult (age over 14 years) medical or surgical patients, who received services at a 500-bed community-based, teaching hospital ("Community Hospital") located in the metropolitan Detroit area of southeastern Michigan. Nearly two-thirds of the Community Hospital patients are drawn from its primary service area, consisting of a single county and several adjacent townships, and the remaining one-third from surrounding rural areas. In addition to Community Hospital are a university hospital and three small private hospitals whose service areas overlap in part with the service area of Community Hospital.

In 1984, active medical staff members at Community Hospital numbered 245, of which approximately 48 percent and 30 percent were credentialed within the departments of medicine and surgery, respectively.

The study sample was drawn from patients hospitalized at Community Hospital during calendar year 1985. For each patient, the earliest 1985 discharge was designated the "index episode." The number of discharges during the 365-day period prior to the patient's index episode represented the "number of prior episodes." The "number of readmissions" was the total number of admissions during the 365-day period subsequent to the index episode.

For an accurate determination of the number of prior episodes and the occurrence and number of readmissions, knowledge of the patient's complete pattern of hospitalization was necessary. Since data available for use in this study pertained only to patient hospitalizations at Community Hospital, not other area hospitals, the possibility existed that counts of prior episodes and readmissions would be underestimated, and in some instances that patients experiencing recurrent hospitalization might be misclassified as single-episode patients. In order to minimize the deleterious effects of incomplete data, the sample was selected to exclude both patients who very probably would use another hospital, and those who were known to have been hospitalized at another acute care facility. Specifically, the sample included only patients whose discharging attending physician at the time of the focal episode practices exclusively at Community Hospital,(1) and patients who resided in the 16 zip code areas proximate to the hospital where the hospital's market penetration was greatest.(2) In addition, a small number of patients (N = 98) known to have been admitted from or discharged to another hospital at the time of their index episode were excluded. After exclusions, the sample consisted of 4,219 patients, representing approximately 33 percent of all adult medical-surgical patients discharged alive from Community Hospital in 1985.

To ascertain whether the selective sampling process had effectively excluded most patients who used other hospitals, aggregate information (patient-specific data were not available due to confidentiality restrictions) was obtained from the Michigan Peer Review Organization regarding the utilization of other hospitals in Michigan by the Medicare patients who constituted nearly 41 percent of the sample. Approximately 7.5 percent of the Medicare patients in the sample were readmitted to another hospital during the 365-day period subsequent to their index discharge. It is reasonable to assume that Medicare patients are generally at greater risk of readmission than are non-Medicare patients, since Medicare is the primary payer for the elderly and severely disabled. Consequently, for the sample as a whole, the percentage of patients readmitted to other hospitals was probably somewhat lower. The selective sampling process appears to have been successful in excluding most patients who used other hospitals.

There were five major sources of data: Community Hospital's patient discharge abstract and accounts receivable data; physician data sets containing descriptive information on the medical staff, such as respective specialty, age, and caseload; information from the hospital's social service records on the posthospital placement of patients in nursing homes or with home health agencies; information regarding the occurrence and timing of fatalities from the death registry of the Michigan Department of Vital Statistics; and estimates of the median income associated with various zip code areas derived from the 1980 United States Census (Donnelley Demographics 1986).

Prior to the regression analyses, the patients in the sample were assigned to one of two groups -- readmitted and not readmitted. The readmitted subsample consisted of those patients having at least one "unplanned" readmission within 365 days of their index episode.(3)

Table 1 provides a list of the predictor and dependent variables used in the study. The dependent variable was "the occurrence of at least one unplanned readmission." There were three groups of predictor variables: patient demographic, socioeconomic, and health status characteristics; measures of the discharging physician's level of experience and performance; and measures of the patient's acute and long-term care service use before, during, and after the index episode.
Table 1: List of Variables
 Patient Socio/Demographic
 Income: median income of area of residence (zip code)
 Payer: Medicare, Medicaid, Blue Cross, other
 Patient Clinical
 MDC: Major diagnostic category (DRG Support Group Ltd. 1983)
 Stage: Stage of illness on an increasing scale of 1 through 4 (see Systemetr
 Inc. 1984)
 Mental: Mental deficiency(*)
 Patient Resource Use
 Prior EPI: prior hospital episodes during previous one year
 LOS: Index episode length of stay
 Nursing Home: Discharge to nursing home after index episode
 Home Health: Discharge with home health services after index episode
 Physician Quality and Efficiency
 MD-Average: Physician age (< 45 years, 45-64, 65-74, 75+)
 Volume: Number of hospital discharges in 1985
 MD-Service: Clinical department membership; coded as 0 for internal
 medicine and 1 for surgery
 Fatality ration: Actual over expected deaths([dagger])
 LOS Ratio: Actual over expected mean length of stay([dagger])
 Readmit: At least one unplanned readmission during one-year period
 subsequent to index episode
(*)For a list of ICD-9-CM codes used to identify functional and mental impairmen
see Corrigan (1987).
([dagger])Measures of physician performance relative to peers within the same cl
department after adjusting for differences in the mix of DRGs.

The Cox proportional hazards regression model was used because the dependent variable was dichotomous and asymmetric, and also because this model provided a graphic depiction of the "time-to-response" (Dixon 1985).


The average and median ages for the sample were 56 and 58 years, respectively, and approximately 53 percent were female. Nearly 62 percent of the sample patients were discharged by physicians from the internal medicine service with the remaining 38 percent from the surgical service. Medicare and Blue Cross were the principal payers for about 41 percent and 35 percent of patients, respectively. The average length of stay was 6.72 days. About 66 percent of the study patients had no Community Hospital discharges during the 365-day period subsequent to their index discharge, while 21 percent had one readmission, about 8 percent had two, and a little over 5 percent had three or more.

In the first phase of the analysis, the sample was divided into two groups -- those who survived (N = 3,823) and those who died during the 365-day follow-up period subsequent to their index hospital episode (N = 396). Surviving and expired patients were analyzed separately for two reasons. First, patients with terminal conditions were hypothesized to exhibit characteristics different from those of patients with long-term debilitating conditions, and to require different approaches to intervention. Second, approximately 40 percent of the terminal sample members died at home or in long-term care facilities and experienced no readmissions to Community Hospital. This represented a sizable number of patients not readmitted who experienced very poor outcomes and, if combined with the surviving subsample, it would have confounded or diluted differences between readmitted and not readmitted patients.

A regression model was fit to the surviving subsample using the major diagnostic category (MDC) of the patient's index episode as a stratification variable (see Table 2). Both increasing index episode length-of-stay and increasing stage of illness were found to be directly associated with increased likelihood of readmission. Patients discharged by an internist were somewhat more likely to be readmitted than those discharged by a surgeon. Prior hospitalizations had the highest significance level of any predictor variable (z = 8.394) with a one-unit increase in the number of prior hospitalizations associated with a 26 percent increase in the log of the readmission rate. Discharge to a nursing home was associated with a 42 percent decrease in the log of the readmission rate at a .01 significance level, while the effect of discharge to a home health agency was not significant.

Due to the presence of multicollinearity, the results pertaining to age, payer, and median income of the patient's area of residence must be interpreted with caution. There appear to be distinct differences in readmission patterns across age groups, with those in the 46 to 65-year age group most likely to be readmitted. However, nearly all patients 65 years and older have Medicare as their principal expected source of payment; consequently, the effect of age on the likelihood of readmission may be reflected in the model as the effect of payer. It is also possible that the inverse effect of income on the likelihood of readmission, which is not quite significant at the .05 level, is reflected in part by the inclusion of the Medicaid payer variable.

To understand better the effects of age and income on the likelihood of readmission, a regression model excluding the payer variables was fit to the data. Although the results are presented elsewhere, increased age was found to be associated with increased likelihood of readmission, and lower socioeconomic status -- as measured by median household income of area of residence -- was also significant predictor of readmission (Corrigan 1987).

The regression model shown in Table 2 was also run for only those patients in the surviving subsample who were less than 65 years of age (Corrigan 1987). In this model, increased age was significantly associated with increased likelihood of readmission, as was having Medicare or Medicaid as a principal expected source of payment. Lower income was weakly associated with increased likelihood of readmission. [Tabular Data 2 Omitted]

A detailed analysis was also conducted on those patients who died during the 365-day follow-up period, and for this subsample there was only one significant predictor variable -- discharge to a nursing home (see Table 3). In this model, the stage of illness was represented as four dummy variables, since preliminary analyses revealed that the relationship between increasing stage of illness and the likelihood of readmission was not linear for the expired subsample (as had been the case with the surviving subsample). Only the coefficient for stage 4 (indicating that the patient had a very poor prognosis at the time of discharge for the index episode) approached significance. [Tabular Data 3 Omitted]

For the expired subsample, discharge to a nursing home was associated with a 67 percent decrease in the log of the readmission rate (alpha -- .01). To determine if the inverse association between nursing home utilization and readmission was consistent across all ages, an age/nursing home interaction term was added to the model. The results, shown elsewhere, indicated that older nursing home residents were less likely to be readmitted than younger nursing home residents (Corrigan 1987).

To provide additional assurance that the study findings were not attributable to chance, a second sample of patients from an earlier time period was selected for purposes of replication. The calendar year 1983 was selected to minimize any overlap in patients between the 1985 study sample and the replication sample. The data set pertaining to the 1983 replication sample was not as complete as that for the 1985 study sample: the discharges were not classified by stage of illness, and information was not available to identify patients who died during the one-year follow-up period. Consequently, prior to replication, it was necessary to fit a new regression model to the entire 1985 study sample (both the surviving and expired subsamples); excluding the stage-of-illness variable; and using as a dependent variable the occurrence or nonoccurrence of a readmission. This revised model was then applied to the 1983 replication sample.

Although the regression analyses run on the study sample and replication sample are presented elsewhere (Corrigan 1987), the study results were found to be robust. As would be expected, the overall model fit to the 1985 sample was more representative of the surviving subsample because they constitute the majority of sample members. In general, increased likelihood of readmission was associated with increased age, male gender, more advanced stage of illness, lower income, increased numbers of prior hospitalization, discharge by an internist, discharge to a community setting, increased index episode length of stay, and younger physicians. When this model was applied to the 1983 replication sample, the size of effects and levels of significance were approximately the same for the age, income, clinical service, prior hospitalizations, subsequent nursing home utilization, subsequent home health utilization, and length of stay covariates. One exception was physician age, which was not a significant predictor in the 1983 replication sample. z-Scores were calculated to determine the significance of the differences in the two samples' beta coefficients, and the only significant difference pertained to the gender covariate, the effect of which was not significant in either of the individual samples.

One traditional step was taken to validate the 1985 study results. The 1985 sample model was used to predict the probability of readmission for each patient in the 1983 replication sample, and the predicted probabilities of readmission were then compared with actual rates of readmission. Obtaining a predicted probability of readmission for each patient in the 1983 replication sample involved both calculation of an estimated readmission function using the patient's actual values for the covariates in the model and conversion of the dependent variable from the log linear function into a probability estimate.

In table 4, the patients in the replication sample were grouped according to the decile of their predicted probability of readmission. For example, according to the 1985 model, 251 of the patients in the replication sample had a probability of readmission of 10 percent or less, and 14 had an estimated probability of over 90 percent. For nearly all subgroups, the actual readmission rate for patients in the subgroup was every close to the predicted probability of readmission for the subgroup. Two deciles (deciles 71-80 and 91-100) in the upper tail of the distribution had actual rates of readmission that fell short of those predicted, but the sample sizes in these upper deciles were quite small. [Tabular Data 4 Omitted]


Hospital episodes in many ways resemble links in a chain. As Donabedian (1973, 84) has pointed out with regard to such a chain: "Each element is, at least to some extent, a cause of the elements that follows, while it is itself caused by the elements that precede it."

This study has identified numerous factors that contribute to patterns of hospital readmission, including patient characteristics such as age, diagnosis, and socioeconomic status; physician characteristics such as experience; and the organization of the acute and long-term care delivery systems. Some of these factors are more amenable to change than others, and are of particular importance in fashioning intervention strategies to alter patterns of hospital readmission.

Generally consistent with other studies reported in the literature, increased age was found to be positively associated with the likelihood of readmission, with the strength of the relationship being less pronounced for the very old (Victor and Vetter 1985; Fethke, Smith, and Johnson 1986; Anderson and Steinberg 1984). Detailed analyses however, revealed that this positive association was present only for patients who survived the one-year follow-up period. Patients in the expired subsample exhibited an inverse relationship between age and readmission, and this relationship was slightly more pronounced for those discharged to nursing homes. Providers are apparently less likely to hospitalize those with very poor prognoses either in deference to patient and family preferences, or in recognition that hospital intervention will have little or no effect on patient outcome. This trend is more pronounced for nursing home patients, probably because nursing homes are more experienced in caring for the terminally ill than are informal caregivers either with or without the assistance of home care services.

Contrary to the very limited information available in the literature (Acheson and Barr 1965; Fethke, Smith, and Johnson 1986), increased socioeconomic status, as measured by the median household income for the patient's area of residence, was also found to be inversely associated with the likelihood of readmission but only for the surviving subsample. Further studies will be needed to understand the direct relationship between socioeconomic status and the likelihood of readmission, as well as any relationships between socioeconomic status and other factors such as informal support systems and the ability to obtain formal community-based support which, in turn, may affect hospitalization. These future studies will need to employ more direct and valid measures of patient income than those used either in this study or in other studies conducted to date.

Advanced stage of illness, discharge by an internist, and increased index episode length of stay were all found to be directly associated with higher rates of readmission for the surviving subsample, but none of these covariates was useful in distinguishing between readmitted and not readmitted patients in the subsample that expired. One explanation for the positive association between increased index episode length of stay and readmission for the surviving subsample is that length of stay is to some extent a measure of severity of illness. It is also possible that some patients who remain in the hospital longer than expected are those who do not have a strong informal support system that will assure aftercare.

Found in at least one other study (Roos et al. 1986b), increased physician age was inversely associated with the likelihood of readmission, at least for the surviving subsample. It would appear that as physicians gain experience, they become somewhat less likely to hospitalize. Another possible explanation is that older physicians refer patients requiring hospitalization to younger colleagues to avoid the rigors of following critically ill inpatients.

In addition to patient socioeconomic and clinical characteristics and physician experience, the likelihood of readmission -- for those in the surviving subsample -- was also found to be associated with the patient's pattern of prior hospitalization. One way to appreciate the size of the effect associated with prior hospitalizations is to (1) specify a set of characteristics for a hypothetical patient; (2) vary the number of prior hospitalizations while keeping all other covariates constant; and (3) compare the various probabilities of readmission associated with each level of prior hospitalization.

For purposes of illustrating the effect of prior hospitalization on members of the surviving subsample, a hypothetical 68-year-old, middle-income, female patient with a nervous system disorder is used. Figure 1 contains three plots of the probability that the hypothetical patient will not be readmitted at any given point in time during the 365-day follow-up period. Each plot assumes a different level of prior hospitalization. The estimates of coefficients used in generating the plots are derived from the model shown in Table 2.

As can be seen from Figure 1, the likelihood of remaining out of the hospital for at least six months is about 78 percent if the hypothetical patient has no prior episodes, as compared with 72 percent and 64 percent for one or two prior episodes, respectively. One year after the index episode, the likelihood that the hypothetical patient with no prior episodes will have experienced a readmission is about 32 percent, while it is 40 percent or greater if the patient has one or more prior hospital episodes.

At least two possible explanations can be given for the positive association between prior hospitalization and the likelihood of readmission. First, patients with prior hospitalization may be more severely ill than those with no prior episodes, and these differences may not be entirely accounted for by the case-mix and stage-of-illness adjustments used in the analyses. Second, patients with a pattern of recidivism may be individuals receiving inadequate formal or informal community-based support.

Subsequent long-term care utilization also has an effect on the likelihood of readmission, but only if institutional services are received. For both the surviving and expired subsamples, the receipt of institutional long-term care services is associated with a significant decrease in the likelihood of readmission. This result is only partially consistent with that of Shapiro, Tate, and Roos (1987) who also identified an inverse relationship between nursing home utilization and hospital readmission, but only for the 75 year and older age group. In this study, long-term care institutionalization was found to have a pronounced effect on readmission for patients both under 65 and over 65 years of age.

The inverse effect of nursing home utilization on hospitalization was found to be somewhat greater for patients in the subsample that expired than for those in the surviving subsample. Figures 2 and 3 present a series of "readmission curves" for a hypothetical patient in the surviving and expired subsamples, respectively. In both examples, the hypothetical patient is a 68-year-old female with a stage 2 nervous system disorder. The projections in Figure 2 are derived from the model shown in Table 2, while the curves in Figure 3 are derived from the model shown in Table 3.

As shown in Figure 2, were a patient with the specified characteristics be discharged to a nursing home, the likelihood of experiencing a hospital readmission within one year would be approximately 28 percent, as compared with 40 percent for a patient discharged to a community setting, either with or without home care services. The difference in the likelihood of readmission between those in an institutional setting and those in a community setting is even more pronounced for the expired subsample. As illustrated in Figure 3, the hypothetical patient's likelihood of readmission within one year is about 30 percent if placed in an institution, as compared to 50 percent when discharged to the community.

One explanation for these findings is that nursing homes, especially skilled nursing facilities, do serve to a significant degree as substitutes for hospital care. Long-term care institutions with trained nursing staffs are capable of managing many acute and chronic illnesses, and are apparently providing a great deal of terminal care. Patients in the community, either with or without home care, must rely to a great extent on lay caregivers; when professionally trained services are required, the hospital is generally the most accessible source.

One limitation of this study is that data were not available to identify the level of care provided to patients discharged to either long-term care facilities or home with home health services. Further research is needed to determine whether there are differences in the likelihood of readmission between patients placed at the skilled nursing versus the custodial levels of care.

The findings of this study challenge many common conceptions of the health care system. The United States health care system has traditionally maintained a fairly strict division between acute care and long-term care in terms of both the delivery and financing of services (Bowlyow 1990). It is evident that certain cost tradeoffs exist between the consumption of hospital services and that of institutional long-term care services. For certain types of patients, such as the terminally ill, hospital consumption is significantly greater if the patient goes to a community setting rather than to an institutional long-term care setting. Further research is needed to identify the cost implications of alternative posthospital placements, and it is imperative that "total costs" be examined -- both acute care and long-term care health services costs. It will also be important in these future studies to assess not only the total health care costs associated with alternative placements, but also patients and family satisfaction and the quality and outcomes of patient care.

The study results also have implications for the financing of hospital services. Per case hospital payment has a tendency to focus clinical management efforts on discrete hospital episodes. A prospective capitation-based payment mechanism provides better incentives for the management of chronically ill patients over long periods of time, especially if the system provides financing and benefits coverage for acute and long-term care services as in a social health maintenance organization (Knickman and McCall 1986).

In addition to the policy issues just mentioned, the study findings also have implications for health care management. The basic predictive model development to identify the characteristics associated with readmission may be adapted for purposes of early identification of patients in need for discharge planning and posthospital case management. The sooner the discharge planning process is underway, the less likely patients will remain in the hospital unnecessarily while awaiting long-term care placement.

It is also apparent that more comprehensive data systems are needed. At present, most health care data bases are specific to a particular part of the health care system, such as inpatient hospital services, ambulatory settings, or nursing homes. Enhancements to these data systems are needed to facilitate the integration and analysis of data regarding all of a patient's health care utilization. In the absence of comprehensive data, it will not be possible to assess the cost and quality tradeoffs associated with alternative patterns of health care utilization.


The authors would like to acknowledge the contributions of Richard Kraft, M.D., Jersey Liang, Ph.D., Jack Wheeler, Ph.D., Robert Wolfe, Ph.D., and the helpful comments of the editor and reviewers of Health Services Research.


(1)Of the 191 medical/surgical physicians at Community Hospital, 109 practiced exclusively at Community Hospital. These 109 physicians accounted for approximately two-thirds of all the hospital's medical and surgical discharges in 1984. (2)Community Hospital accounted for at least 25 percent of all hospitalizations for residents of each of the 16 zip code areas. For the 16 zip codes combined, the commitment index for Community Hospital was 37 percent, and these zip codes accounted for 55 percent of all Community Hospital discharges. (3)A planned readmission was on that (1) occurred within 30 days after hospitalization, involving an arteriogram, a myelogram, or an endoscopy; and (2) involved treatment for the same problem associated with the index episode. A list of the "pairs" of ICD-9-CM diagnosis and/or procedure codes used to identify planned readmissions may be found in Corrigan (1987).


Acheson, E. D., and A. Barr. "Multiple Spells of In-Patient Treatment in a Calendar Year." British Journal of Preventive Social Medicine 19 (1965): 182-91. Anderson, G. F., and E. P. Steinberg. "Hospital Readmissions in the Medicare Population." New England Journal of Medicine 311, no. 21 (1984): 1349-53. Bowlyow, J. E. "Acute and Long-Term Care Linkages: A Literature Review." Medical Care Review 47, no. 1 (1990): 75-103. Corrigan, J. M. Patient and Health System Characteristics Associated with the Utilization of Hospital Services and Long Term Care Services Subsequent to Hospitalization. Ann Arbor, MI: University Microfilms, Dissertation Abstracts, 1987. Dixon, W. J., ed. BMDP Statistical Software. Berkeley, CA: University of California Press, 1985. Donabedian, A. Aspects of Medical Care Administration. Cambridge, MA: Harvard University Press, 1973. Donnelley Demographics. Donnelley Marketing Information Services. Stanford, CT: Donnelley Demographics, 1986. DRG Support Group, Ltd. The Revised ICD-9-CM Diagnosis-Related Groups. New Haven, CT: DRG Support Group, Ltd., 1983. Fethke, C. C., I. M. Smith, and N. Johnson. "Risk Factors Affecting Readmission of the Elderly into the Health Care System." Medical Care 24, no. 5 (1986): 429-37. Graham, H., and B. Livesley. "Can Readmissions to a Geriatric Medical Unit Be Prevented." The Lancet 1 (February 1983): 404-406. Holloway, J. J., J. W. Thomas, and L. Shapiro. "Clinical and Sociodemographic Risk Factors for Readmission of Medicare Beneficiaries." Health Care Financing Review 10, no. 1 (1988): 27-36. Holloway, J. J., S. V. Medendorp, and J. Bromberg. "Risk Factors for Early Readmission among Veterans." Health Services Research 25, no. 1, Part II (1990): 213-37. Knickman, J. R., and N. McCall. "A Prepaid Managed Approach to Long-Term Care." Health Affairs 5, no. 1 (1986): 90-104. McCall, N., and H. S. Wai. "An Analysis of the Use of Medicare Services by the Continuously Enrolled Aged." Medical Care 21, no. 6 (1983): 567-85. Riley, G., and J. Lubitz. "Outcomes of Surgery in the Medicare Aged Population: Rehospitalization after Surgery." Health Care Financing Review 8, no. 1 (1986): 23-34. Roos, L. L., S. M. Cageorge, N. P. Roos, and R. Danzinger. "Centralization, Certification, and Monitoring: Readmissions and Complications after Surgery." Medical Care 24, no. 11 (1986a): 1044-66. Roos, N. P., G. Flowerdew, A. Wajda, and R. B. Tate. "Variations in Physician's Hospitalization Practices: A Population-based Study n Manitoba, Canada." American Journal of Public Health 76, no. (1986b): 45-51. Roos, N. P., and E. Shapiro. "The Manitoba Longitudinal Study of Aging: Preliminary Findings on Health Care Utilization by the Elderly." Medical Care 19, no. 6 (1981): 644-57. Shapiro, E., R. B. Tate, and N. P. Roos. "Do Nursing Homes Reduce Hospital Use?" Medical Care 25, no. 1 (1987): 1-8. Systemetrics, Inc. Disease Staging: A Clinically Based Approach to Measurement of Disease Severity. Vols. I-V. Santa Barbara, CA: Systemetrics, Inc., 1983, 1984. Victor, C. R., and N. J. Vetter. "The Early Readmission of the Elderly to Hospital." Age and Ageing 14 (1985)": 37-42. Zook, C. J. S. F. Savickis, and F. D. Moore. "Repeated Hospitalization for the Same Disease: A Multiplier of National Health Costs." Milbank Memorial Fund Quarterly 58, no. 3 (1980): 454-71. Janet M. Corrigan and James B. Martin This research was supported in part by a doctoral student award from the Michigan Health Care Education and Research Foundation, and through the generosity of an anonymous hospital. Janet M. Corrigan, Ph.D., Director, Planning and Development, National Committee for Quality Assurance, 1730 Rhode Island Avenue, N.W., Suite 307, Washington, DC 20036. James B. Martin, Ph.D., is Associate Professor of Health Services Management and Policy at the School of Public Health, University of Michigan, Ann Arbor. This article, submitted to Health Services Research on September 5, 1990, was revised and accepted for publication on May 16, 1991.
COPYRIGHT 1992 Health Research and Educational Trust
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1992 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Corrigan, Janet M.; Martin, James B.
Publication:Health Services Research
Date:Apr 1, 1992
Previous Article:Volume responses to exogenous changes in Medicare's payment policies.
Next Article:The effects of utilization review on hospital use and expenditures: a covariance analysis.

Terms of use | Privacy policy | Copyright © 2022 Farlex, Inc. | Feedback | For webmasters |