Utilization patterns of cohorts of elderly clients: a structural equation model

Utilization patterns of cohorts of elderly clients: a structural equation model

A.Y. Ellencweig

The elderly account for a relatively large share of total use of health care resources in Canada. For example, in British Columbia (BC) in 1985-1986, those over 65 years of age used about 60 percent of all patient days in acute, extended, and rehabilitation care hospitals (Evans, Barer, Hertzman, et al. 1989). While they represented less than 12 percent of the population, this same group received medical services representing about 23 percent of total fee-for-service payments to physicians (Barer, Pulcins, Evans, et al. 1988). The extensive utilization of health services by the elderly population and the availability of large data sets in Canada inspired extensive utilization studies on longitudinal health care use. In BC, researchers have analyzed the first five years of the care of persons admitted to the BC Long Term Care (LTC) program (Stark, Gutman, and McCashin 1982; Stark, et al. 1984; Stark and Gutman 1986; Lane, Uyeno, Stark, et al. 1985; Lane, Uyeno, Stark, et al. 1987). The study extended into a far more comprehensive data set describing the first-year health care experience of a 1981-1982 admission cohort into the BC LTC program (Stark et al. 1987).

The preliminary analysis of the new data set involved study of total utilization (Ellencweig, Pagliccia, McCashin et al. 1990; Ellencweig, Stark, Pagliccia, et al. 1990) and use of forecasting methodology for the estimation of future utilization (Krueger, Ellencweig, Uyeno, et al. 1989). While valuable for estimating overall utilization, those techniques are not sufficiently sensitive either to discern trends in the utilization of specific services, or to assess the effects of the use of certain services on utilization of other services and whether these effects are determined by the users’ characteristics such as age and gender. For example, how do GP contacts affect referrals to specialists? Or, how does the performance of diagnostic procedures affect hospital use? The evaluation of these interactions requires a path model. Path models have been used in health services research on the elderly when influences of health services use were examined for predisposing factors, health status, and social support networks (Andersen and Laake 1983; Wan 1987). However, those models did not show how cohort characteristics affected the relationships between individual services and levels of utilization for each service.

The current study has three objectives:

1. To develop a basic model capable of describing the utilization of health services by individuals before their admission to the LTC program;

2. To assess deviations from the basic model, attributable to gender and age, in patterns of utilization by clients before admission to the LTC program; and

3. To describe changes and differences from the basic model in utilization patterns that occurred during the year after admission to the LTC program, attributable to admission, age, and gender.


The study population is the 1981-1982 admission cohort of the BC LTC program. The data for each person in the study are taken from three computer databases maintained by the provincial government for payment and administrative purposes. These three files are:

1. The LTC file. A record of each client at the time of admission to the LTC program. Variables include the age and gender of the client; geographic location; provider types grouped into home care, facility care, and day care only; and the level of care.

2. The Medical Services Plan (MSP) file. The vehicle by which physicians are paid; it contains claims initiated by physicians for services (visits, surgery, etc.) provided to patients.

3. The hospital file. The instrument that records service information for each period of hospitalization.

Each of the files is maintained independently of the others, and extensive computer processing allowed the development of unique client identifiers that permitted utilization data for each individual to be linked throughout the study period. The absence of deductibles or copayment, coupled with the physicians’ financial incentive to claim for services, limited the likelihood of underreporting. Since British Columbia residents are reimbursed for medical care that occurs outside as well as inside the province, these events were also recorded.

Utilization data for each individual were collected by type of service. Information was included on most service variables that had been made available through the linked data set. The few exceptions were those utilization variables that were either poorly defined (such as “other types of physician service”) or included too many service categories (such as minor surgical procedures performed in an ambulatory setting by GPs as well as specialists). The database used in this article includes age, gender, and 15 health services utilization variables counting the units of service for each person. Variable definitions as adapted from the Ministry of Health guidelines (Medical Services Plan 1985) are given in the Appendix. All of the information is available for the year before and the year after admission to LTC. The variable “age” was divided into two: patients less than 80 years of age, and those 80 or older. Because health status is a major determinant in utilization by the elderly, our population includes only those who were in “personal care” (a relatively low need for medical care) at admission; the total comprised 57 percent of clients at all levels of care, or 2,906 individuals.

Given the highly skewed utilization data resulting from the large number of LTC clients with no physician contacts, a first attempt to achieve a more symmetric distribution was made by excluding from the cohort those clients without any type of GP contacts in both periods. Nonusers of GP services in both periods have not contributed to utilization changes over time. This is also true for specialist and hospital use, since in British Columbia further contacts can only be made after an initial contact with the GP. The final study population consisted of 2,198 individuals for the year before admission and 2,188 individuals for the year after admission. This is over 75.6 percent of the total admissions to personal care. (More information on the BC LTC program and the study population can be obtained from the authors.)


A model capable of describing the complex LTC health services utilization system will have to take into account the interdependency of the health services. For example, not only are GP hospital visits dependent on hospital length of stay, but hospital length of stay is dependent on specialist consultations and hospital consultations are, in turn, dependent on GP hospital visits. Variables that have this type of relationship are said to be jointly dependent. In this study 10 of the 15 variables are considered to be jointly dependent.

As a result of the interdependency between health services variables, additional effects on the jointly dependent variables are of interest. For instance, specialist consultations may have a direct effect from GP hospital visits and also from GP office visits, but there may also be an effect from GP office visits indirectly through GP hospital visits. The total effect is the sum of the direct plus the indirect effects.

Table 1 shows the relationships between the ten jointly dependent variables considered in our model. Thus, for example, the first row indicates that a hospital visit is dependent on (or is a function of) hospital stay and intensive care days, and the last row indicates that major surgery is a function of all the other jointly dependent variables. The asterisks represent the coefficients to be estimated. In addition, each jointly dependent variable is assumed to be a function of the remaining independent variables: GP office, home, nursing home, night, and emergency visits.

This is the proposed conceptual model based on the British Columbia health services system. It is believed that GP visits are independent of other services except for GP hospital visits, which is dependent on hospital and intensive care days. It is also believed that diagnostic and laboratory tests, and major surgery are dependent on most other services utilization, while specialist office visits are dependent only on specialist consultations and home visits (aside from all the independent variables). We reason that other relationships may be less obvious but still justifiable. For example, the length of hospital stay is assumed to be dependent on specialist consultations but not on specialist hospital visits; rather, the number of specialist hospital visits is a function of hospital stay.

The two characteristics of the health services utilization process heretofore mentioned (interdependency and presence of indirect effects) can be analyzed in a linear structural equation model. The analyses were performed using LISREL-Linear Structural Relationships (a system of multivariate regression equations for one- or two-group analysis; for the analysis of two groups LISREL simultaneously fits a model to the two groups; the difference in parameter estimates indicates the extent of the difference between the groups [Joreskog and Sorbom 1986]).


First, an effort to achieve normality and variance reduction was attempted by performing a log-transformation of the values of all variables plus one. Even though the marginal normality so achieved does not guarantee joint normality, the improvement in the distribution will certainly be more conducive to the proper use of parametric statistical techniques (Gnanadesikan 1977).

The first step of the analysis was to fit the basic model to the utilization data for the admission cohort (all clients with at least one GP contact) during the year before admission. Once the basic model was fitted, it was used as the reference point for eight additional analyses of different groups: (1) male and female clients, before admission; (2) males less than 80 and 80 or older, before admission; (3) females less than 80 and 80 or older, before admission; (4) all clients, year before admission; (5) males less than 80 before and after admission; (6) males 80 or older before and after admission; (7) females less than 80 before and after admission; and (8) females 80 or older before and after admission. In each case the goodness of fit was established by the TABULAR DATA OMITTED LISREL goodness-of-fit indicator as well as by the normed fit index delta (Bentler and Bonett 1988) and the comparative fit index (Bentler 1990).


Table 2 summarizes the evaluation of the nine final models in each analysis based on various goodness-of-fit indexes. Large p-values and fit indexes close to 1 generally indicate good model fits. The null model represented the most restrictive situation. In the case of two-group analysis the null model implied the same parameter estimates across groups; in the case of single-group analysis (years before or after admission), the null model implied independence between hospital and nonhospital utilization.



For the sake of clarity, in the following presentation we speak of:

* Regular (nonemergency) GP contacts (office, home, and nursing home visits);

* Emergency GP contacts (emergency room and night visits);

* Specialist contacts (consultations, home and office visits);

* Diagnostic/surgical procedures (diagnostic, laboratory and surgical procedures); and

* Hospital use (hospital stay, intensive care days, and specialist/GP hospital visits).


Basic Model

Table 3 shows the estimated regression coefficients and total effects associated with the jointly dependent variables and the independent variables. Only statistically significant (t [is greater than] 2) coefficients are shown. First we consider the direct effects.

Specialist consultations, laboratory tests, and diagnostic procedures (with coefficients of 0.271, 0.341, and 0.12, respectively) are considerably dependent on the number of GP office visits. The number of GP home visits has only a relatively weak (although significant) effect on utilization variables such as specialist consultations, hospital stay, laboratory tests, and surgery procedures, whereas nursing home visits have no significant effect on any of the utilization variables considered in the model.

Visits in the emergency room affect most other utilization variables, in particular, the number of GP hospital visits, specialist consultations, and hospital stay: the coefficients are 0.127, 0.245, and 0.193, respectively. Physician night visits have a pattern similar to that of physician visits to the emergency room; they have no effect on specialist hospital visits and diagnostic procedures. The smaller coefficients also indicate a weaker effect.

In conclusion, GP contacts associated with regular community care (GP office, home, and nursing home visits) do not significantly affect hospital-related use (except for a marginal relationship between visits to the physician’s office and hospital stay). On the other hand, GP contacts are strongly correlated to hospital use in a crisis situation (GP emergency and night visits).

The coefficients of the jointly dependent utilization variables indicate that the largest effect by far is that of length of stay on GP hospital visit (0.762). Length of stay also has a considerable bearing on specialist hospital visits (0.316), diagnostic procedures (0.559), and surgical procedures (0.363). Specialist consultations also have a significant effect on several utilization variables: specialist office visits (0.41), specialist hospital visits (0.290), and diagnostic procedures (0.371). Finally, surgical procedures are also dependent on diagnostic procedures (0.488). Other meaningful relationships can be seen in Table 3. Some negative relationships are worth noting: GP hospital visits were found to have a negative effect on specialist consultations (-0.186), diagnostic procedures (-0.262), and surgical procedures (-0.152).

Next, in order to assess the overall impact on the utilization variables, we consider the total effects (bottom values in Table 3). For most relationships, the total effects are not very different from the direct effects. However, the total effect of GP emergency room visits on length of stay is 0.303 compared to 0.193 direct effect. Further, the total effect of specialist consultations on surgical procedures (0.441) is far greater than the direct effect (0.140). Similarly, the total effect of specialist consultation on specialist hospital visits and diagnostic procedures is considerably larger than the direct effect. On the other hand, the total effect of hospital stay on each of the variables shown in Table 3 is smaller than (or equal to) the direct effect alone. For example, the coefficient for total effects of hospital stay on diagnostic procedures is 0.236, as compared to the direct effect of 0.559; this mitigated effect is the result of a negative indirect effect of length of stay on diagnostic procedures via intensive care days and specialist hospital visits.

The proposed LISREL model fits the data for the year before admission, and therefore it was adopted as the basic model for health services utilization of a cohort of clients first admitted to the LTC program in BC. Before attempting to compare the utilization data for the year before admission with data of the same cohort for the year after admission, the basic model was used to examine the gender and age subgroups of the cohort for the year before admission. (Details on parameter estimates other than for the basic model can be requested from the authors.)

Male and Female Clients

The model indicates that male and female clients have a statistically different utilization pattern. The final model reveals that this difference is structural, as shown by the reversal of the relationship between specialist consultations and office visits, suggesting that specialist consultations are a function of specialist office visits for male clients, as opposed to the relationship in the basic model. The pattern of utilization by female clients is consistent with the basic model.

Although other differences between males and females are small and spread among all variables rather than being concentrated on only a few, males show a somewhat different pattern. This is more noticeable in the smaller effect of specialist home visits and in larger effects of specialist consultations on the dependent variables for male clients as opposed to female clients.

Male Clients under Age 80 and Males 80 or Older

The next analysis was made between younger male clients ([is less than]80) and older male clients (80[is greater than or equal to]) for the year before admission. The two groups are not significantly different when a structural change is introduced in the basic model; that is, specialist consultations are now a function of specialist office visits (the relationship reversed from the basic model above), with the associated total effect coefficient now quite large (0.406). This is consistent with the model for male clients in the previous subsection. Aside from the structural change for male clients, GP home and nursing home visits generate insignificant overall health services activity. As well, GP emergency utilization has no significant effect on specialist office visits and on diagnostic procedures.

The total effects coefficient estimates for the jointly dependent variables have, with some exceptions, the same order of magnitude as in the basic model. Specialist home and office visits have a larger effect on diagnostic and laboratory procedures, respectively; specialist office visits also have a larger effect on hospital days. On the other hand, hospital stay has a smaller effect on GP hospital visits.

Female Clients under Age 80 and Females 80 or Older

The last analysis for the LTC utilization data during the year before admission is for female clients less than 80 and female clients 80[is greater than or equal to]. A clear trend distinguishes utilization by younger female clients from that of their older counterparts. Hospital visits by a GP have a larger negative effect on diagnostic procedures and surgical procedures for older females than for those who are younger. This means that for older female clients, fewer diagnostic and therapeutic activities are performed as a result of GP hospital visits. Also, specialist consultations and specialist office visits have smaller (or not significant) effects on diagnostic and hospital-related variables for older females. In turn, diagnostic procedures have larger total effects on laboratory and surgical procedures for younger female clients. However, once admitted, older females require more hospital care than younger ones, as seen from hospital stay affecting GP visit and intensive care days more strongly.


Total Cohort

Correlations between utilization variables for the year after admission tend to be generally higher than for the year before. This is the case for night and emergency visits by a GP, laboratory procedures, intensive care days, and specialist office visits. One exception is diagnostic procedures that have generally smaller correlations, especially with specialist consultations, hospital stay, and intensive care days. The sign of the coefficients is unchanged except for GP visits to a nursing home.

The basic LISREL model was fitted to the utilization data for the year after admission to the LTC program. It was found that the basic model cannot describe in its totality the health care utilization pattern during the year after ([[Chi].sup.2] = 14.29, d.f. = 2, p = .001). However, subsequent tests on parameters’ invariance across periods, by age and gender, produced good model fits except for male clients under age 80.


Younger Males under Age 80

The data do not appear to support the same structural model for younger male clients in the two periods ([[Chi].sup.2] = 20.48, d.f. = 4, p = .000). In the year after, two structural equations–the specialist home visits equation and the equation involving intensive care days–seemed to be explained better (i.e., they have larger coefficients of determination) by the utilization variables. A closer examination of the differences in the values of coefficients shows clear trends that are more marked in the total effects.

For referrals to specialists and for community-based procedures ordered by a GP, the pattern for the year after admission was more intensive than that for the year before. Specialists’ visits, on the other hand, resulted in a decrease in referred activities in the year after admission. Hospital-related activities, too, decreased in the year after admission in comparison to the year before.

Male Clients 80 or Older

The analysis of male clients 80[is greater than or equal to] in the two periods includes specialist consultations as a function of specialist office visits for the year before admission, and the reverse relationship for the year after admission. In the year after admission, the estimated total effects are larger than in the year before for relationships that depend on GP regular contact in the community (i.e., office and home visits by a GP). Further, the effect of GP office visits on diagnostic procedures is significant only in the year after. For emergency contacts, the effects of GP emergency visits on GP hospital visits, specialist consultations, and hospital days are larger in the year before; but the effect of emergency visits on intensive care days and specialist hospital visits is significant only in the year after.

Specialist activity (i.e., consultations, and office and home visits) generates more hospital services usage in the year before admission than in the year after. Specialist consultations affect hospital days and specialist hospital visits and, similarly, specialist home visits affect intensive care days and specialist hospital visits more strongly before admission to the program. For clients admitted to an acute care unit, specialist activity has a larger effect on the intensity of hospital-related procedures in the year after admission.

Females under Age 80

A good model fit is achieved for the group of younger females. The effect of GP office visits on specialist consultations, specialist office visits, and diagnostic and laboratory procedures is smaller in the year after admission. Generally, GP visits have a larger effect on hospital services utilization (i.e., hospital visits and hospital days). In turn, specialist services and hospital-related utilization had smaller effects in the year after admission.

Females 80 or Older

Aside from having specialist consultations as a function of specialist office visits for the year after admission (reversed relationship from the year before admission), this group has a new relationship for the two periods: GP hospital visits as a function of specialist office visit.

Analysis of the total effects coefficients reveals that more specialist referrals are generated by each regular visit to a GP in the year after admission. However, GP emergency visits yielded larger effects on hospital use as well as on most specialist services in the year before admission. Regular contacts with specialists resulted in more diagnostic tests and surgical procedures in the year before admission than in the year after. However, specialist contacts resulted in more hospital use in the year after.


In this article we studied the health services utilization patterns of a group of new admissions to personal care in the LTC program for 1981-1982 in British Columbia. Because utilization of health care services by the elderly depends primarily on their health status, we restricted our analysis to those in relatively good health. We have used a multi-equational model that (a) reflects the jointly related nature of health services utilization variables, and (b) provides a global view of the health services utilization “system” quantified in terms of total effects. This model has confirmed some findings from previous research, indicating the suitability of the approach, and at the same time providing additional insights regarding patterns of utilization. The model also provides provoking implications for decision makers.

The Entire Cohort

The analysis of the model for the entire cohort in the year before admission (the “basic” model) shows that, for regular GP contacts, the physician seeing the patient is likely to order a laboratory test and slightly less likely to seek a specialist consultation. In turn, specialist consultations generally result in more intensive diagnostic screening and more hospital stays than do regular GP contacts. From a cost-oriented perspective, these findings may imply higher costs for care rendered by specialists than for care given by generalists (Eisenberg and Nicklin 1981; Rosenblatt, Cherwin, and Schneeweiss 1982).

In the case of GP emergency contacts, hospital stay is the more likely outcome. In turn, for hospital-based activities, longer stays in acute care units lead to more extensive utilization of resources–physician contacts, diagnostic tests, and use of intensive care.

While the analysis of direct effects conforms to previous findings, the analysis of the total effects provides a more realistic view of the health care delivery process. For instance, while some direct effects are small (or even negative), the total effects are positive and proportionally larger. The most striking case is the much larger positive effect that GP emergency room visits have on hospital specialist visits (3.6 times larger), on GP hospital visits (3 times larger), and on hospital stay (1.5 times larger). This may suggest that although an emergency room visit shows minor direct effects on hospital-related activity, it factually translates into more utilization when the whole health care network is set in motion. This happens simply because patients seen in the ER are more likely to be more severely disabled. Beland, Lemay, Philibert, et al. (1991) reached a similar conclusion: that the use of emergency services entails extensive use of hospital resources.

While general utilization patterns in the basic model are quite informative, the analysis of subgroups of clients and changes over time can be even more effective in explaining patterns of use and interrelationships between health services variables.

We now discuss the seven different cases of group analyses shown in the results section.

Age Effects

The pattern of health services utilization of younger and older male clients during the year before admission to LTC is characterized by a more active hospital use and a decrease in diagnostic/surgical procedures for the older males. Older female clients, on the other hand, show a marked decrease both in hospital use and diagnostic/surgical procedures as compared to the younger female clients. It is interesting that larger negative total effects of GP hospital visits on diagnostic and surgical procedures were found for both older males and females: age does appear to reveal different patterns of utilization. Indeed, previous research has shown this to be common among older clients (Wolinsky, Arnold, and Nallapati 1988).

The reduced level of utilization can be explained by three of the six hypotheses suggested by Wolinsky et al. (1988). First, as previously noted, older clients seem more likely to have available informal support by other members of the family (Wolinsky, Arnold, and Nallapati 1988; Brody 1985). Second, the older old are likely to reflect later stages of chronicity than are younger patients. By that time, treatment regimens are stabilized, and therefore, presumably, less medical intervention is required. Third, it is possible that health professionals share a feeling that they cannot do much to improve older clients’ conditions and, therefore, they discourage the latter from making as many follow-up visits or periodically scheduled visits as they had before.

Gender Effects

As shown in several studies summarized by Wolinsky and Arnold (1988), utilization by males and females differs for several types of health care services, including doctor visits and hospital lengths of stay. In our analysis, the overall effect of the differences in utilization is reflected by structurally different models. The feature that characterizes the structural difference in the utilization pattern between male and female clients is the opposite direction in the relationship between specialist office visits and consultations. For females, a visit to a specialist’s office appears to be determined by (or caused by) a specialist consultation; this is to say that specialist consultations by females generate a sizable number of specialist visits. For males, the opposite is true; in a similar proportion, visits to a specialist generate a sizable number of specialist consultations. An explanation for this phenomenon is not obvious except by noticing that, in the case of female clients, specialist consultations (for a second opinion) materialize in the utilization of all the other health services considered in this study, except those provided by GPs, with a generally greater effect than in the case of male clients. For males, the effect of a specialist consultation on the other utilization variables is smaller.

When both gender and age are considered, older female clients tend to have a less extensive utilization profile than younger clients, as indicated by the generally smaller total effects. These findings conform to earlier findings from the Health Interview Survey in the United States as reported by Wolinsky and Arnold (1988). The authors show that for females of Anglo origin, the percentage of physician contacts drops by 3 percent between ages 75-84 and 85 and older, while the comparable drop for males is less than 1 percent. For patients of African American origin, the difference is even larger: a 4 percent drop in utilization for females compared to a 2 percent increase for males. Why do older women have a more reduced volume of utilization than men? Perhaps because informal substitution of health care services by family caregivers is more effective for women (Wolinsky, Arnold, and Nallapati 1988). Perhaps doctors often value treatment for older males differently than for older females. This may make them more likely to recommend fewer follow-up visits and greater intervals between periodical visits for their older female patients. This hypothesis is strengthened by some evidence in the literature that doctors tend to use more extensive resources to treat male patients when the clinical factors are similar (Ayanian and Epstein 1991; Steingart, Packer, and Hamm 1991).

By and large, the most informative results in pairwise analysis are those showing how a given cohort changes its utilization pattern before and after admission to LTC. Indeed, the utilization before and after admission was not identical for any one of the four pairwise analysis. Three of the four cohorts reflected similar utilization behavior, but the cohort of younger males appears to have a drastically different behavior in the postadmission period.

Given the interdependency of the utilization of health services, changes over time cannot be seen as simple downward or upward shifts for the whole cohort; rather they appear as trade-offs between different services as a result of other services utilized by each age/gender group. Therefore, perhaps there is no single answer when asked whether the utilization during the year after admission is lower or higher than that of the previous period. The answer may depend on which services are being referred to and which services have an effect on other services.

We discuss the three major service utilization changes:

Change in Specialist Services Utilization. Specialist services utilization tends to be larger in the year after admission as a result of GP regular visits, except for younger female clients, where it is generally lower as a result of other specialist services. This is consistent with a pattern of utilization typical for clients admitted to the personal care level in the LTC program. For those clients the same health condition that induced admission to LTC might have persisted during the year after, requiring follow-up visits by specialists but not other types of services.

Change in Hospital Services Utilization. Hospital services utilization tends to be about the same for younger males during the year after admission. This confirms previous findings (Ellencweig, Pagliccia, McCashin, et al. 1990) on a similar group of clients who did not show a significant difference in length of stay between the two periods. But no causal relationships were established then. Clearly, the hospital services utilization patterns for the other study groups do not show consistent trends. Where factors that generate utilization are considered, prior hospital use is highly correlated to increased hospital use in the year after admission to LTC.

Changes in Diagnostic/Surgical Procedures Utilization. Only as a result of specialist and hospital services utilization is there a decrease in diagnostic and surgical procedures. Thus, specialist and hospital service utilization can be seen as an alternative to surgery. The impact of GP services, on the contrary, is toward an increase of diagnostic/surgical procedures. Because visits to GPs are more likely to reflect new episodes of care than visits to specialists, the increase in diagnostic or even surgical procedures resulting from the former is straightforward. Diagnostic/surgical procedures seem to generate a smaller number of like procedures.

In view of our results, it would be simplistic to propose that the LTC program either produces or does not produce an overall decrease in health care utilization. Our model suggests less deterministic (but more qualified) conclusions: after admission to LTC, more general utilization, with some exceptions, is observed as a result of GP visits (both regular and emergency); and, again with some exceptions, the pattern of utilization appears to be more reduced in the year after as a result of specialist utilization. Certainly, when we consider the dynamic processes of the health care system in a single structural model, the exceptions qualify the conclusions and simultaneously point to the complexity of the relationships between health utilization variables.

One limitation implicit in the data set is the use of discrete time data instead of continuous time event history data. That is, client utilization data are aggregated in one-year time units and not as utilization events in their chronological sequence of occurrence. Time lags between services within the year are not taken into account. This is a common practice in similar studies where the client (and not the event) is the unit of observation. We believe this limitation does not completely invalmdate the causal relationships of our model due to the referral system operating in British Columbia: specialist contacts can occur only as a result of a GP referral; diagnostic/surgical procedures can occur only as a result of a GP or specialist contact.

Our model provides substantial benefits beyond the mere study of changes in utilization after entering the LTC program. It gives a much better insight than the traditional utilization studies on the intricate network of health services use. The strength of the model lies in its ability to account for joint dependency and indirect effects which are frequently neglected by analysts of health services. In our analysis, we have repeatedly shown that the discussion of indirect effects is important from both a theoretical and a practical perspective, as indirect effects may change considerably the magnitude and even the direction of the relationships caused by the direct effects alone.


We would like to thank an anonymous referee for helpful comments on an earlier draft, but we are responsible for remaining errors.


Andersen, A., and P. Laake. “A Causal Model for Physician Utilization: Analysis of Norwegian Data.” Medical Care 21, no. 3 (March 1983): 266-78.

Ayanian, J. Z., and A. M. Epstein. “Differences in the Use of Procedures between Women and Men Hospitalized for Coronary Heart Disease.” New England Journal of Medicine 325, no. 4 (July 1991): 221-25.

Barer, M. L., I. R. Pulcins, R. G. Evans, C. Hertzman, J. Lomas, and G. M. Anderson. Diagnosing Senescence: The Medicalization of British Columbia’s Elderly. Vancouver: Health Policy Research Unit, University of British Columbia, 1988.

Beland, F., A. Lemay, L. Philibert, B. Maheux, and G. Gravel. “Elderly Patients’ Use of Hospital Based Emergency Services.” Medical Care 29, no. 5 (May 1991): 408-18.

Bentler, P. M., and D. G. Bonett. “Significance Tests of Goodness of Fit in the Analysis of Covariance Structures.” Psychological Bulletin 88, no. 3 (1988): 588-606.

Bentler, P.M. “Comparative Fit Index in Structural Models.” Psychological Bulletin 107, no. 2 (1990): 238-46.

Brody, E. M. “Parent Care as Normative Family Stress.” The Gerontologist 25, no. 1 (1985): 26-29.

Eisenberg, J. M., and D. Nicklin. “Use of Diagnostic Services by Physicians in Community Practice.” Medical Care 19, no. 3 (March 1981): 297-309.

Ellencweig, A. Y., N. Pagliccia, B. McCashin, A. Tourigny, and A. J. Stark. “Utilization Patterns of Clients Admitted or Assessed but Not Admitted to a Long-term Care Program-Characteristics and Differences.” Canadian Journal on Aging 9, no. 4 (Winter 1990): 356-70.

Ellencweig, A. Y., A. J. Stark, N. Pagliccia, B. McCashin, and A. Tourigny. “The Effect of Admission to a Long-Term Care Program on Utilization of Services by the Elderly in British Columbia.” European Journal of Epidemiology 6, no. 2 (June 1990): 175-83.

Evans, R. G., M. L. Barer, C. Hertzman, G. M. Anderson, I. R. Pulcins, and J. Lomas. “The Long Good-Bye: The Great Transformation of the British Columbia Hospital System.” Health Services Research 24, no. 4 (1989): 435-59.

Gnanadesikan, R. Methods for Statistical Data Analysis of Multivariate Observations. New York: John Wiley and Sons, 1977.

Joreskog, K. Y., and D. Sorbom. LISREL Analysis of Linear Structural Relationships by the Method of Maximum Likelihood, Version VI: User’s Guide. Sweden: University of Uppsala Department of Statistics, 1986.

Krueger, H., A. Y. Ellencweig, D. Uyeno, B. McCashin, and N. Pagliccia. Physician Utilization before and after Entering a Long-Term Care Program: An Application of Markov Modelling. Vancouver: University of British Columbia, Health Policy Research Unit, 1989.

Lane, D., D. Uyeno, A. Stark, E. Kliewer, and G. Gutman. “Forecasting Demand for Long-Term Care Services.” Health Services Research 20, no. 4 (1985): 435-60.

Lane, D., D. Uyeno, A. Stark, G. Gutman, and B. McCashin. “Forecasting Client Transitions in British Columbia’s Long-Term Care Program.” Health Services Research 22, no. 5 (1987): 671-706.

Medical Services Plan. Guide to the Submission of Physicians Accounts for Insured Services. Victoria, British Columbia: Ministry of Health, 1985.

Rosenblatt, R., D. Cherwin, and R. Schneeweiss, “The Structure and Content of Family Practice: Current Status and Future Trends.” Journal of Family Practice 15, no. 4 (1982): 681-722.

Stark, A. J., G. M. Gutman, and B. McCashin. “Acute-Care Hospitalizations and Long-Term Care: An Examination of Transfers. “Journal of the American Geriatric Society 30, no. 8 (1982): 509-15.

Stark, A. J., E. Kliewer., G. M. Gutman, and B. McCashin. “Placement Changes in Long-Term Care: Three Years’ Experience.” American Journal of Public Health 74, no. 5 (1984): 459-63.

Stark, A. J., and G. M. Gutman. “Client Transfers in Long-Term Care: Five Years’ Experience.” American Journal of Public Health 76, no. 11 (November 1986): 1312-16.

Stark, A. J., N. Pagliccia, B. McCashin, and C. Jackson. “Services Utilization by Elderly Home and Facility-Based Clients Admitted to Continuing Care in British Columbia.” Presented at the 115th Annual Meeting of the American Public Health Association, New Orleans, 1987.

Steingart, R. M., M. Packer, and P. Hamm. “Sex Differences in the Management of Coronary Artery Disease.” New England Journal of Medicine 325, no. 4 (July 1991): 226-30.

Wan, T. “Functionally Disabled Elderly: Health Status, Social Support and Use of Health Services.” Research on Aging 9, no. 1 (1987): 61-78.

Wolinsky, F. D., and C. L. Arnold. “A Different Perspective on Health and Health Services Utilization.” Annual Review of Gerontology and Geriatrics 8 (1988): 71-101.

Wolinsky, F. D., C. L. Arnold, and I. V. Nallapati. “Explaining the Role of Physician Utilization among the Oldest Old.” Medical Care 26, no. 6 (1988): 544-53.

Appendix: Definitions of Service Utilization Variables Used in the LISREL Utilization Model

General Practitioner Services

Office Visit A visit to the physician’s office during office hours

Home Visit A day visit at the client’s home; call placed between 0800 and 1800 hours

Nursing Home Visit Similar to home visit if client resides in a nursing home

Night Visit An evening or night visit at the client’s home or in a nursing home; call placed between 1800 and 0800 hours

Hospital Visit A physician visit of client admitted to acute care

Emergency Visit A physician visit of client in an emergency room

Specialist Services

Office Visits A visit to the specialist’s office

Home Visits A specialist visit at the client’s home

Hospital Visit A specialist visit of client admitted to acute care

Specialist Consultation Defined as a request by a doctor for a second opinion on a case that he or she has examined, encountering some difficulties; care returned to the referring doctor

Other Services

Diagnostic Tests All diagnostic procedures including laboratory, x-ray, electrocardiogram, and nuclear medicine; procedural episodes only; physicians’ opinions listed in addition

Laboratory Tests Laboratory tests; only the actual cost of the laboratory included

Major Surgery Major operative procedures performed in an acute care hospital (minor procedures and outpatient surgery excluded.) (Since it was not possible to assign specific surgery to individual physicians it is assumed that those procedures were all performed by specialists.)

Hospital Stay Total number of hospital days (acute care) during a specified period of time

Intensive Care Days Total number of days spent at an intensive care unit during a specified period of time

Source: Medical Services Plan. Guide to the Submission of Physicians Accounts for Insured Services. Victoria, Province of British Columbia: Ministry of Health, 1985.

COPYRIGHT 1994 American College of Healthcare Executives

COPYRIGHT 2004 Gale Group