This paper assessed trends in utilization patterns for cancer hospitalizations, and determinants of those trends, during a decade of health care reform. Using data drawn from a 10% sample of British Columbia health service users over the years 1991–1998, we asked two questions. First, do hospitalization rates display discontinuities consistent with the timing and/or intent of health care reform? Second, are there significant associations involving area of residence, income quintile or hospitalization year, or interactions of area, quintile and year, with hospitalizations?
Trends in hospitalization rates
Findings indicated that all separations rates did not display any discontinuity consistent with a system responding to abrupt policy change. Instead, steady reductions in cancer separations rates were observed over the decade, in agreement with previous studies of general hospital utilization rates following health system reforms. Specific cancer hospitalizations, with the exception of colorectal cancer, which is stationary, also displayed consistent decline over time. Hence, health care reforms through the 1990s did not precipitate sudden alterations in general cancer hospitalization trends.
First separations were examined as proxy incidence, on the proposition that an individual with cancer will be treated at a hospital at least once during the course of the disease. Incident cases indicate demand for services over time. As incidence, first separations are unlikely to be affected by health policy, since policy changes observed do not occur within the time frame necessary to influence cancer disease development, a process that takes many decades. Policy changes may influence first separations where such changes allow increased access to extant or new screening procedures. In addition, first separations may be influenced by changes in technology and treatment patterns, for example where a change in technology allows early detection, as was the case for prostate cancer through the 1980s and 1990s.
Unlike all separations, all cancers first separations rates display a discontinuity, and consequently two line segments; the joinpoint occurs just after the start of 1995. Prior to the joinpoint, the slope of the line is significantly negative; following, the slope is significantly positive. Lung cancer first separations also contain a discontinuity during 1995. Again, the first segment displays a significantly negative slope; post-1995 the slope is not significantly different from zero. In contrast, colorectal, breast, and prostate cancer first separations rates are all adequately described by a single line. Of the three, only the prostate cancer first separations slope is significantly (negatively) different from zero.
We use other separations to examine hospitalization trends without reference to first separations. All cancers' other separations exhibit a joinpoint slightly following that of first separations, and mirror the first separations pattern. That is, while the second segment of first separations has positive slope, other separations' second segment has negative slope. The consistently declining all separations pattern, then, is a result of other separations' rate decrease despite first separations' rate increase.
In contrast, other separations for each specific cancer require only one regression line to describe the trend over time, and each exhibits a significantly negative slope. As noted above, first separations should reflect demand and not changes in policy or treatment. By extension, all and particularly other separations should reflect these latter changes. Only all cancers other separations display a discontinuity, in 1995, two years following the initial implementation of administrative restructuring in 1993.
If hospitalizations for cancer follow a characteristic pattern, we would expect first and additional, or other, separations rates to vary in parallel over time. Furthermore, this should hold for all cancers (an aggregate of many different treatment patterns), and with reference to specific cancers. If the trends do not vary in parallel, then there will be intervening factors that explain the disparities. From Table 2, directions of slopes agree only for prostate cancer, where both first and other separations decline over the study period. However, for all cancers, as well as for lung, colorectal and breast cancer, first and other separations yield disparate results. For example, breast cancer first separations slope is not different from zero, while that for breast cancer other separations is significantly negative. At least for all cancers, lung, colorectal and breast cancer, it would appear that some event or events during 1991 through 1998, other than first separations (incidence), affected other separations rates.
There are several sources of explanation for changes in these rates. In addition to health policy, they may include changes in incidence, modifications in detection and treatment, or alterations in treatment location (e.g. a shift from hospital to another community location). We cannot deal directly with effects of modifications in detection or treatment, including treatment location, using these data. However, we can discuss the likelihood of changes in incidence affecting overall decline in hospital utilization over the study period.
Declines in all and other separations rates may be a result of reduced demand, that is, either reduced absolute frequencies or standardized rates of incident cancers. However, British Columbia Cancer Agency figures [51] indicate an increasing absolute count of incident cases from 1970 through 2000. Similarly, previous BCLHD analysis (paper in manuscript) showed increases in frequency of absolute counts of first separations in British Columbia over the 1990s. In addition, National Cancer Institute of Canada statistics [43] also note increasing absolute numbers of cancer cases as the Canadian population ages. Thus, decreased absolute numbers of disease cases cannot be the source of decreasing hospitalizations.
What about age-sex adjusted incidence rates? If these rates are decreasing, it would suggest that age-standardized demand for hospitalization will decrease, without influence from health care reform or other developments. Canadian cancer statistics [43] indicate, with some exceptions, that incidence rates for cancer in Canada were stable or declined during the 1990s. For example, all cancers age-standardized incidence rates for men ranged between 4.46 and 4.94 per thousand from 1991–1998; for women rates were between 3.35–3.45 per thousand. Calculations for average annual percentage change (AAPC) in age-standardized incidence rates, among the specific cancers examined here for 1991–1998, a period precisely contiguous with the present study, indicate the following. First, incidence AAPC in all cancers among men was significantly negative (-1.2%), whereas for women there was no significant change (0.1%). For lung cancer, both men and women experienced significant change, but in opposite directions (for men, -2.3%, and for women, +1.6%). Colorectal cancer exhibited a significant downward shift for men (-0.6%) and no significant change for women (-0.7%). AAPCs for both breast cancer and prostate cancer were not different from zero (0.2% and -1.5% respectively).
These AAPCs in incidence rates, used as measures of changing annual demand for services, can be compared to the estimated annual percent change (EAPC) in hospitalizations calculated here and presented in Table 3. All separations EAPCs for all cancers and each specific cancer, other than colorectal cancer, were significantly and consistently negative, whereas EAPCs for first separations were not significantly different from zero for all or any specific cancer except prostate. EAPCs for other separations for all categories, except all cancers first line segment, were significantly negative. Again, separations rates appear to be influenced by factors in addition to simple cancer incidence. These factors may include alterations in policy, treatment patterns and locations.
Finally, rates may vary over time as a result of changes that shift procedures from in-patient to out-patient admission (or vice versa). However, data analyzed here include both in- and out-patient separations, suggesting such changes would not be reflected in this analysis.
Potential determinants of hospitalization
Table 4 presents results from the multivariate analyses of area of residence, income quintile and separation year, with all, first and other cancer hospitalizations counts (offset by population). Consider the first column of Table 4, describing results for all cancers in aggregate. How can these be interpreted?
For all cancer hospitalizations, income quintile does not figure as an important player, whether first, other or all hospitalizations are considered. Area, on the other hand, while non-significant for first separation, matters when either all or other hospitalizations are assessed. Detailed examination of the GENMOD-GEE results indicates that the more rural areas (CHSSs) experience more events (hospitalizations) than more urban areas (RHBs) for both all and other separations. This result is in agreement with prior analysis (data not shown), indicating elevated all separations age-sex adjusted rates in Upper Island – Central Coast, a CHSS, in contrast to rates from the province and two RHBs. Separation year is significantly associated with each of first, other and all separations. Detailed parameter estimates indicate all and other separations exhibit elevated events earlier in the study period. These results are consistent with observed trends in hospital utilization, indicating declines in recent years. First separations parameter estimates, in contrast, indicate fewer separations earlier in the time period. In other words, all and other separations decrease over time despite rising first separations counts.
Which variables associate significantly with specific cancers? With regard to area of residence, the more rural CHSS populations experience elevated hospitalization levels over urban RHB populations for lung and prostate cancers, and lung, colorectal and prostate cancers, for all and other separations respectively. There is no association of area with first separations for any specific cancer.
Income quintile is significant only for all separations for breast cancer. Here, pairwise contrast of quintiles shows higher incomes (Q3 – Q5) form a consistent cluster as do lower incomes (Q1 and Q2). However, and curiously, Q1 is not significantly different from Q5. Where significant, parameter estimates indicate fewer hospitalizations amongst lower income quintiles. We calculated rates for income clusters defined as Q1 – Q2 vs. Q3 – Q5. Figures 6 through 8 depict age adjusted breast cancer separations rates for all, first and other separations respectively, for these clusters. Rates were calculated relative to at-risk groups, that is, rates for cluster 1 (Q1 and Q2) were calculated with the denominator as age structured population by income cluster 1. All rates were standardized to the 1991 BC female population. Income clusters all exhibit zero slope. Therefore, to the extent that there are differences across income, as is the case for all separations, these differences were maintained over the decade.
Separation year matters for all separations in prostate cancer only; lung and colorectal cancer first separations; and lung, breast and prostate cancer other separations. For prostate cancer all and other separations, parameter estimates tend to be larger in earlier years. The direction of this association is to be expected since incident prostate cancer cases fell from historic highs in the late 1980s and early 1990s, due to increased detection of cancer following TURP (trans-urethral resection of the prostate) and early detection due to prostate specific antigen (PSA) screening tests [43]. Lung and breast cancer other separations' parameter estimates, where significant, are also greater in earlier years of the study period. However, this pattern is reversed for first lung and colorectal cancer separations. That is, and again where significant, parameter estimates are lower in earlier relative to later years. This trend reflects increasing first separation absolute counts.
Finally, are there any interactions between area of residence, income quintile and separation year? We are interested in whether there are differences in association between areas of residence or income quintiles and hospitalization rates over time that might indicate changing equity of access to care. However, our findings indicate no significant associations for either area of residence or income quintile with first separations in aggregate or specific cancers. In addition, no significant effects were observed for area-year, income-year, or area-income-year interactions for any cancer(s) or separations category. Consequently, we infer there has been no significant change over the study period in ability to obtain access to hospital cancer care relative to area of residence or income level in British Columbia.
Conclusions
Did health care reform unequivocally affect aggregate and specific cancer hospitalizations during the study period, from 1991 through 1998? The answer would appear to be no. Our findings indicate no consistent abrupt changes with respect to all separations that can be associated temporally with health policy reform. Insofar as reforms were designed in response to existing trends, which dictated reductions in hospitalizations, then they acted in concert with those trends. Reforms may have permitted all and other hospitalizations to decline despite relatively constant first separations rates, by creating or buttressing community resources that obviate hospitalization during cancer treatment. Shifting locations for treatment and recovery may influence steady declines in hospitalizations, particularly where health policy directs provision of community care for those with chronic diseases. Direct evidence for this possibility is lacking within data presented here; this will be the subject of future analyses.
Did health care reform alter equity of access to hospital care in different region and income groups? First separation, considered by itself without reference to incidence, is not significantly associated with area of residence or income quintile, nor is there any observed interaction for first, all or other separations between area and year, income quintile and year, or area, income and year. Insofar as these data are concerned there has been no significant change over the study period in access to care across groups, and these findings concur with previous reports indicating no change consequent on health care reform.
If first separation is indeed a valid marker for incidence, then based on these data British Columbia residents do not appear to be developing cancer differentially by socio-economic status or region of residence. However, we also note that an association between SES and cancer incidence has been observed in the Canadian context. For example, Gorey et al. [19], using Toronto data from the Ontario Cancer Registry, observed significantly greater lung and colon cancer incidence among areas with lower SES, and greater breast and prostate cancer incidence among areas with higher SES. Either BCLHD first separations are not a reasonable proxy for incidence, or the BCLHD sample analysed here failed to detect incidence differentials among income groups, or the Toronto differentials do not hold in British Columbia. Until these alternatives can be differentiated, we are reluctant to conclude that equity in access to care has been achieved or maintained.
This study has several limitations. BCLHD data analysed here were requested originally for a study primarily examining regional variation in general service utilization during health reform. The provincial sample represents 10% of BC residents who used the health care system from 1990–1991 through 1998–1999. As such, the sample is a longitudinal, nine-year, 10% sample, rather than a single annual 10% sample followed longitudinally, or nine annual 10% cross-sectional samples. Further, the 10% sample with respect to cancer hospitalizations is small for some areas, particularly when subdivided by specific cancer, age, sex, income quintile and separation year. We will be examining a larger BCLHD sample, requested specifically for this purpose, to establish whether results presented here are consistent. Finally with respect to the sample, reforms were planned and variously implemented throughout the 1990s. Data analysed here extend only to 1998. It is possible that analysis beyond 1998 may show altered hospitalization trends.
With respect to first separation acting as a marker for incidence, at the moment the suggestion is a logical conjecture supported indirectly rather than empirically verified. We are proceeding to directly test the relationship. In addition, and as indicated above, the income measures that were used are based on 1996 Census enumeration areas, and therefore the analysis is ecological rather than individual. We note that literature on cancer and SES provides a rationale for using area income measures, given evidence that community SES is significantly associated with stage-at-diagnosis and survival (for example, in Canada [24], but see [19]; in the United States, [52–54]). Finally, it is beyond the scope of the present analysis to evaluate effects of changes in cancer treatment on hospital and community care utilization. Future analyses, with data specifically requested for this purpose, will examine treatment patterns and cancer service utilization, equity and health outcomes among the British Columbia population.
In 2001, following a change in government, British Columbia again restructured its health care system [55]. Studies examining health system changes during the 1990s will be important not just to evaluate effects of those reforms on utilization and population health status, but also to provide a baseline against which to evaluate this most recent round of health system reorganization.