Chapter IX Hospitalization Key Words: * Admissions in ESRD hospitalization * Hospitalization by regions * Dialysis hospitalization * Dialysis unit profit status * Standardized hospitalization ratio Hospitalization rates reflect many aspects of ESRD therapy. Among the most important are the frequency and duration of reported hospitalizations, both of which are significantly affected by the level of patient morbidity. Unfortunately, other influential (but unrelated) factors include the health insurance system and individual patient needs. Consequently, hospitalization data are subject to numerous sources of variability, and tend to be imperfectly reported at both patient and aggregate levels. Despite these faults, such data allow the USRDS to provide reasonably objective characterizations of the morbidity experience in the ESRD population. The source of hospitalization data for this chapter are the Medicare billing records contained in the HCFA standard analysis files (SAF; see Chapters I and XIII) for the years 1991-1995. This represents a two-year increment in hospitalization data over that reported 1996 ADR; the latter covered data collected through 1993. The majority of analyses presented here will be based on 1993-1995 data only; 1991 and 1992 data are used primarily in an empirical investigation of time trends. While many of the problems which plagued previous hospitalization analyses have been eliminated (e.g., 1995 ADR; see also the discussion in the 1996 ADR), some limitations still remain. For example, patients in their first 18-21 months of Medicare eligibility who have Medicare as their secondary insurance payer may have their hospital stays covered by another source first, with Medicare being the secondary payer. This may result in hospitalization profiles for a minority of patients that are incomplete during the first 18 to 21 months. The new procedures used to generate the SAF hospital records do screen patients to determine if Medicare is likely to be a secondary payer by requiring that patients included in the analysis have at least one Medicare-paid dialysis bill prior to the start of the study period. This process helps to reduce bias due to missed hospitalizations, but does not completely alleviate the problem. Such limitations, while somewhat less important for comparisons made in the aggregate, do inhibit investigations into the use of serial hospitalizations as a measure of illness severity, longitudinal comorbidity, quality of life, and monetary cost of ESRD, at least on a per-patient basis. The first part of this chapter summarizes the hospitalization experience of incident and prevalent dialysis patients through total hospital admissions and hospital days. These analyses are done on a "per calendar year" basis, and consequently do not adjust for the fact that members of the study cohort are at risk for hospitalization for differing periods of time. The second part of this chapter describes the hospitalization rates of incident and prevalent dialysis patients, defined as the total number of admissions per year at risk for hospitalization. Only current dialysis patients, defined as those patients who have been on dialysis for at least 60 days by their study entry date (to be defined), are included in these analyses. Prior to the 1996 ADR, patients who had been previously transplanted were automatically excluded; this is no longer the case. That is, previously transplanted dialysis patients who satisfy the study entry criteria are now included in all analyses. However, patients who died of AIDS are retroactively excluded from all analyses to avoid any biases resulting from the variation in prevalence from this disease. The analyses in the last section of this chapter utilize a recently developed comparison measure based on standardized first hospital admission rates (Strawderman et al., 1996). This standardized first hospitalization ratio (SHR) is an adaptation of the standardized mortality ratio (SMR) methodology described in Chapters V and XIII. Using these standardized rates, we compare the SHR and SMR among dialysis units within the United States Department of the Census’s nine census regions. We also investigate the relationship between SHRs and SMRs at the dialysis unit level, and study trends both among the various facility types (i.e. free-standing/hospital and profit/not-for-profit status) and over time. The reasons for concentrating on first hospitalizations rather than total hospitalizations are explained in some detail later in this chapter. In short, the SHR and rates computed based on total admissions provide essentially the same summary information; however, the advantage of the former over the latter is that statistical inference based on the SHR is much more straightforward. Summaries by age, race, sex, diabetic status and modality group are given in the reference tables (Section H), and apply to data collected between 1993-1995. There are some new additions to the hospitalization reference tables this year. First, a breakdown of SHRs by state for the 1993-1995 is given in Table H.2. A second addition to this years’ reference tables is a 3 year (1993-1995, Table H.5) and one year (1995, Table H.6) summary of total hospital admissions by diagnosis (DRG) code. We note that most of the analyses in this chapter are meant to be descriptive, not definitive, and thus serve only to generate hypotheses for detailed study. Hypothesis testing, confidence intervals, etc. are reserved for future, more in-depth analyses of these data. Trends in Hospitalization Both the yearly number of hospital admissions and days per patient are important measures in the study of hospitalization in dialysis patients. Figures IX-1 and IX-2 describe the distribution of hospital admissions and days for patients prevalent on January 1 or incident during 1995, by patient age under or over 65. Please note that this represents a two year increment over the summary statistics reported in the 1996 ADR. ------------------------------------------------------------------------ Figure IX-1 [Image] Figure IX-1. Percentages of dialysis patients with the given number of hospital admissions in a single year, by age, 1995. Mean and median are also denoted. Includes all Medicare dialysis patients. Please note that this figure shows counts of hospital admissions for the entire year rather than admissions per year at risk. As a result, the numbers here are lower than those for total admissions per year at risk. Source: Special Analysis. ------------------------------------------------------------------------ It should also be noted that these figures are based on the number of events per calendar year rather than the number per year at risk. Given that time at risk is often much less than one year, the number of hospital admissions and days per year at risk is greater than those numbers shown here. ------------------------------------------------------------------------ Figure IX-2 [Image] Figure IX-2. Percentages of dialysis patients with the given number of total days hospitalized in a single calendar year by age, 1995. Mean and median are also denoted. Includes all Medicare dialysis patients. Please note that this figure shows counts of days in the hospital for the entire year rather than days per year at risk. The numbers here are lower than those for days per year at risk. Source: Special Analysis. ------------------------------------------------------------------------ We see that the majority of patients have few days and admissions; there is also a long tail extending to the right, and as expected the distribution of days is skewed more so than total admissions. More patients have zero admissions and zero days than any other number. Figure IX-1 shows that approximately 80.5 (81.7) percent of the patients in the 65+ (under 65) age groups had fewer than 3 hospital admissions, while 96 (97) percent respectively had 5 or fewer. The median number of admissions is 1 per year in both groups, while the mean number of admissions in the older and younger age groups are 1.4 and 1.3 respectively. These are unchanged from the mean and median number of admissions for 1993, which were previously reported in the 1996 ADR. However, a comparison of the percentages in each admission category between 1993 and 1995 (not shown) indicates the distribution of admissions is shifting slowly to the left. In terms of hospital days, we see in Figure IX-2 that 39 percent and 48 percent of the patients in the older and younger age groups had zero hospital days; in 1993, 34 percent and 42 percent respectively had zero days. Just over 11 percent of 65+ year olds had 30 or more days in the hospital, while patients under the age of 65 had only 9.5 percent. This represents a drop since 1993 in the younger age group. In 1993, the median and mean for hospital days were respectively 5 and 12 for patients age 65+ and 3 and 11.5 for patients less than 65. In 1995, however, the median and mean for hospital days were respectively only 3 and 11.4 for patients age 65+ and 1 and 9.6 for patients less than 65. In Table IX-1, the median and mean total number of hospitalization admissions and days is given for 1991-1995. In conjunction with the above results, these yearly comparisons show that while the total number of admissions has decreased only slightly, the amount of time spent in the hospital has decreased significantly. ------------------------------------------------------------------------ Table IX-1 [Image] ------------------------------------------------------------------------ In general, the trends observed here are similar to those reported in the 1995 and 1996 ADR’s; younger patients tend to have less hospitalization, measured on either scale, and the distribution of both total hospital admissions and days continues to shift slowly to the left, indicating an overall decrease in hospitalization. The apparent decrease in hospitalization is consistent with the current overall national trend of decreasing hospitalization, which itself reflects efforts to contain overall health care costs. Crude Hospitalization Rates Based on Total Admissions The patient populations under study in Figures IX-1 and IX-2 include dialysis patients prevalent on January 1, 1993, or incident after that date until December 31, 1995, and hence includes patients having different lengths of followup. These descriptive analyses do not account for this fact, and therefore we computed an aggregated rate per year at risk for hospitalization, defined as the ratio of total hospital admissions or days to the total time at risk. In order to stabilize the estimated rates, we pooled data for the period 1993-1995, calculated yearly totals for the number of hospital admissions and years at risk for hospitalization, and use a weighted average of these numbers to obtain the overall rate. These rates are then computed for various age, race, sex, and modality groups within the ESRD population in order to compare the hospitalization experience among groups. Patient Eligibility and the "60-Day Rule" To provide the most accurate snapshot of the hospitalization experience among dialysis patients, we restricted the population of patients used in calculating the rates. The rates are based on data pooled yearly over 1993-1995, meaning that the observed numbers of admissions and corresponding risk time are calculated yearly. To contribute data to the rate calculation in a given year, a patient must be on dialysis for at least 60 days prior to their study entry date. This is done primarily to ensure that adequate Medicare payment data (and hence information on hospitalizations) are available. Incident patients, classified as those patients whose 90th day of ESRD falls between January 1 and December 31, automatically satisfy this condition; correspondingly, the study entry date for each incident patient is set to their 90th day of ESRD. Prevalent patients have their eligibility determined as of January 1; any previously-transplanted patient currently on dialysis having a failed transplant within 60 days prior to January 1 is excluded. The latter is done under the presumption that such patients may contribute hospitalizations early during the followup period which are due to the transplant failure and not dialysis-related complications. Patients who start followup or die during a year are at risk for only a portion of the year, and eligible patients who are transplanted during a given year have their at-risk period censored 3 days prior to the date of transplantation. Finally, patients who switch modalities during the year are assigned to the new modality at the start of the next year. As discussed earlier, all AIDS patients are retroactively excluded. Collectively, these criteria will subsequently be referred to as the "60-Day Rule," and help to ensure that the maximum amount of reliable information on the hospitalization of dialysis patients is included in the calculations. Rates Based on Total Admissions The hospitalization rates summarized in Figures IX-3 and IX-4 are calculated under the "60-Day Rule." Our comparisons here are restricted to rates based on total admissions since corresponding rates based on hospital days are highly correlated. In addition, admission rates tend to exhibit greater stability than those based on days, in part due to more accurate recording and the fact that the distribution of hospital admissions tends to have a shorter right-hand tail (see Figures IX-1 and IX-2). For the admissions rates, the years at risk for hospitalization for each patient are determined by subtracting the time actually spent in the hospital from the total time on observation. ------------------------------------------------------------------------ Figure IX-3 [Image] Figure IX-3. Number of hospital admissions per patient year at risk for ESRD patients by method of treatment for each age cohort, 1993-1995. Patients are grouped by treatment type based on the treatment method they were using on January 1 of each year. Those patients who switched modalities during a year are reclassified at the beginning of the following year. Medicare patients only. Source: Special Analysis. ------------------------------------------------------------------------ This calculation explicitly accounts for the fact that one is not at risk for a new hospitalization while already in the hospital. In counting total admissions, hospitalizations that overlap or occur without any days between discharge and the subsequent admission are combined into a singe hospitalization spanning from the admission of the first to the discharge of the last; this reduces the number of hospitalizations by about 10 percent. This methodology is different from that of other researchers (HCFA 1994). Rates for hospital days broken down by age, race, sex, and modality groups may be found in Reference Table H.4 and use the full years at risk, including time spent in the hospital. ------------------------------------------------------------------------ Figure IX-4 [Image] Figure IX-4. Number of hospital admissions per patient year at risk by sex and race for each age cohort, 1993-1995. Medicare dialysis patients only. Source: Special Analysis. ------------------------------------------------------------------------ In order to properly interpret the total admission rates (TAR), and in particular the standard errors reported in Table H.4, some assumptions must be made. Standard approaches to statistical inference on rates, and in particular the calculation of standard errors, are typically based on the Poisson distribution (Ross, 1983). In the case of the total hospitalization rates reported here, this requires one to assume that the time between successive hospitalizations within each individual are uncorrelated, and that, within each individual, the distribution of the number of admissions over a fixed time period follows a Poisson distribution. Thus, for example, these assumptions imply that the chance that a patient enters the hospital within two days after being discharged is essentially the same, say, as it is for the two day period following 6 months after discharge. When such an assumption is reasonable, the aggregate number of hospitalizations in the same time period should follow a Poisson distribution, at least approximately (Ross, 1983). When these assumptions are violated, calculation of standard errors for hospitalization rates based upon the Poisson assumption is misleading. In addition, the interpretation of the numerical value of the rate itself may be affected. It was shown in the 1996 ADR (Figure VIII-3) that the Poisson assumption is in fact unlikely to hold for the data collected in 1993; while not shown here, this problem persists for both 1994 and 1995 data. One reason for this problem may be related to a larger than average risk for future hospitalizations of patients who had more than one admission during the year. Consequently the TARs reported here should be considered solely as descriptive summaries and should be interpreted with caution. In the next section, an alternative method for computing hospital admission rates is presented, and largely avoids these problems by restricting the calculations to "first admissions" only. Figure IX-3 shows the total admission rates by modality, stratified upon age. The rates generally increase with age for each treatment modality, most noticeably in the aggregate group "All ESRD" (note: this group includes all ESRD patients, not just those dialysis patients selected under the 60-day rule). Hospitalization rates among CAPD patients are slightly higher than for hemodialysis patients in each age group, with the exception of those patients in the 65 and over age group, where the two are equal. This is consistent with trends seen in the 1996 ADR. Interestingly though, CAPD hospitalization in the 0-19 age group went down while that in the 65+ plus age group went up. In contrast, hemodialysis rates either stayed constant or dropped. It was reported last year that hospitalization rates for CAPD patients have in recent years been steadily falling while those for hemodialysis patients have remained relatively stable. Habach et al (1995) reported such comparisons for 1988 through 1990. Suggested reasons for reduced hospitalizations among CAPD patients included changes in the frequency of switching between modalities and differences in age distributions among the two treatment groups. The differences in hospitalization reported here and in the 1996 ADR are smaller than those found in Habach et al (1995), perhaps due to improvements in connection devices thereby reducing peritonitis risk or increased outpatient treatments of peritonitis in CAPD patients. The fact that there is no significant decrease in the CAPD hospitalization rate and in some cases a slight increase over those rates reported in the 1996 ADR perhaps suggests that the rate of hospitalization among CAPD patients is beginning to stabilize. Actual differences between rates for CAPD and hemodialysis patients might be larger than are reported because we have used an "intent to treat" assignment of dialysis modality. That is, hospitalizations for patients who switch in the middle of the calendar year are not attributed to their new modality until January 1 of the following year. The USRDS has previously shown that CAPD patients switch to hemodialysis approximately three times as often as hemodialysis patients switch to CAPD (1995 ADR). If hemodialysis patients tend to incur less hospitalizations, reported rates for CAPD patients may be biased downward while rates for hemodialysis patients are likely to be biased upward (although much less so) since a switch to CAPD likely increases the hospitalization rate. ------------------------------------------------------------------------ Figure IX-5 [Image] Figure IX-5. Distribution of standardized first hospital admissions ratio (SHR) and mortality ratio (SMR) for each of the nine geographic census regions, 1993-1995. Ratios are standardized for sex, race, age and diabetic status. Rates are calculated by dividing the actual number of events (first admissions for hospitalization and deaths for mortality) for each dialysis unit by the expected number of events for that unit. Expected events numbers are calculated for each facility based on national data so that rates can be compared from region to region. Ratios are given for the 10th percentile, the 25th percentile, the median, the 75th percentile and the 90th percentile. National average is 1.05. Units with fewer than twenty expected first admissions in the three year period are excluded. Source: Special Analysis. ------------------------------------------------------------------------ Figure IX-4 shows rates by age, race, and sex. The trends seen here are nearly identical to those reported in the 1996 ADR. Insofar as the rates are themselves interpretable, it is reasonable to infer from these results that among ESRD dialysis patients aged 20+, hospitalization rates tend to rise with age within each race/sex category, females tend to have higher rates of hospitalization than do males, and that Asians have significantly lower hospitalization rates in general. It is important to note that the rates for patients aged 0-19 are, relative to the other age groups, based on rather small sample sizes. This is especially true for Asians and Native Americans, and to a lesser extent Blacks and Whites. Consequently, small changes in either the number of admissions or time at risk may result in large changes in the calculated rates, and patterns observed these groups over short time periods may simply be due to rate instability rather than any real difference. Furthermore, the rates for the youngest age group are also likely to be affected by patient selection: Given the high transplantation rate in this group, those who have never been transplanted are likely to be less healthy and hence higher hospitalization rates. Standardized Hospitalization Ratio (SHR) Methods The standardized mortality ratio (SMR) is a ratio of the observed number of deaths for a given patient study group divided by expected number of deaths for that patient study group based on national death rates. Wolfe et al (1992) use the published USRDS national ESRD mortality rates given in deaths-per-patient-year by age, race, and diagnosis group. The USRDS has since updated this methodology and now uses a more sophisticated model-based procedure to compute ESRD mortality rates at the national level. A more precise description of the methodology is given in Chapters V and XIII. These national ESRD mortality rates can then be used, for example, to compare mortality rates among dialysis facilities with different patient mix characteristics by simply computing the ratio of the observed number of deaths to the expected number within each group, the latter being adjusted for differences in age, race, and diagnosis. The expected number of deaths within a group is determined by multiplying the total patient-years observed within each age-race-diagnosis category by the corresponding national rate, and then summing over all of the categories. An observed SMR larger (smaller) than 1.0 denotes potentially a higher (lower) mortality rate than the national ESRD norm. The SMR is subject to random variation, however, and thus should be interpreted cautiously and not without some evaluation of statistical significance. Further discussion of such matters can be found in Wolfe et al (1992) and also Wolfe (1994). The USRDS produces national ESRD hospitalization rates in a similar manner to the mortality rates. Hence, for patients eligible under the "60-day rule", we can calculate a standardized hospitalization ratio (SHR) using the rates in Table H.1. In calculating the SHR, we restrict our attention to the first hospitalization event for each individual. That is, within a given year, only the first hospitalization event for an individual is counted. Correspondingly, the risk time for that individual is defined as the days from entry until a first hospitalization, a censoring event, or December 31 occurs. Censoring events are death and transplant; a patient’s risk period is truncated 3 days prior to transplant in order to avoid attributing the transplant-related hospitalization to the observed count. National first hospitalization rates are obtained for 248 patient subgroups defined by age (16 groups), race (4 groups), sex (2 groups), and diabetes (2 groups) in an essentially identical manner to the SMR; see Chapters V and XIII for in-depth discussion. In short, a log-linear Poisson regression model is used to smooth the observed national first hospitalization rates; the resulting rates represent weighted averages of the observed and model-predicted rates, with the observed rate being weighted more heavily for larger patient subgroups. The advantage of this approach is that the observed rates in some patient subgroups are rather variable from year to year due to small numbers of patients; taking a weighted average with the model-predicted rate stabilizes (i.e., reduces the variability of) the resulting rate across time. The rationale for considering only the first hospitalization event is explained in detail in Strawderman et al (1996). To summarize that discussion, let us first consider the SMR. In calculating the SMR, each individual makes at most one contribution to the observed death count. These independent contributions to the numerator of the SMR are in effect what allows inference about the SMR to be based on the Poisson distribution; this includes the calculation of standard errors. Hoem (1987) provides an excellent discussion. In contrast to the SMR, the contributions to the observed number of total hospital admissions, which would comprise the numerator of any standardized total admissions rate (STAR), are not necessarily independent of each other. That is, there are many individuals who can contribute more than one hospitalization event in the same time period, and both intuition and empirical evidence suggest that the times between successive hospitalization events are not independent within an individual. If true, this necessarily implies that the total observed count cannot be distributed as a Poisson random variable (Ross, 1983), and in turn effects both the calculation of standard errors and evaluation of statistical significance of the observed STAR at, say, the facility level. Indeed, as discussed earlier (see also Figure VIII-3, 1996 ADR), the total observed hospitalization count does in fact not appear to follow a Poisson distribution, and consequently such issues are of some relevance here. The use of only the first hospitalization event for each individual helps to ensure that contributions to the observed count remain independent of each other, and consequently that methods similar to the SMR may be used to draw inference about the calculated rates. ------------------------------------------------------------------------ Figure IX-6 [Image] Figure IX-6. Map of United States indicating the nine Census regions. Shading indicates those regions with median standard first hospital admissions ratios greater than 1.05 in the three year period, 1993-1995. ------------------------------------------------------------------------ As pointed out in Strawderman et al (1996), the SHR reflects the useful information found in an appropriately defined STAR. For example, at the dialysis facility level, a low SHR necessarily indicates a low overall admissions rate; obviously, if there are few first admissions, there can be few total admissions (unless a few patients at the facility have particularly chronic hospitalization patterns). Correspondingly, a high SHR indicates that many more patients at the facility are entering the hospital than at the national level. Compared to the STAR, the SHR is less sensitive to the level of comorbidity at the patient level, and more sensitive to the scope (or distribution) of comorbidity at the facility. From the point of view of facility evaluation, the latter seems more relevant. Finally, in view of the fact that approximately 65 percent of the patients had 1 or fewer admissions ( see Figure IX-1), the results of the analyses to be presented will be similar to any proper analysis of rates based on total admissions. To obtain the SHR for a specific dialysis unit in a specific year, the total number of first hospital admissions for each eligible patient treated during that time period is divided by the expected number of first hospitalizations. The expected number of first hospitalizations is calculated similarly to the expected number of deaths used in calculating the SMR. Specifically, the observed patient-years at risk for hospitalization in that unit is sub-divided by age, race, sex, and diagnosis, multiplied by the corresponding national rate for those groups, and then summed up over all groups to obtain the total expected number of first hospitalizations in that unit for that year. This produces standardized first hospitalization rates, adjusted for age, race, sex, and diagnosis, that share a similar interpretation to the adjusted SMR. That is, values of the SHR larger than 1.0 indicate first hospitalization rates above the national norm while values below 1.0 denote lower rates. Analyses by Region The SHRs computed for the analyses of this section are based on data obtained for 1993-1995. The "60-Day Rule" is used to determine patient eligibility for each of these years. The SHRs analyzed here are based on total first admissions and corresponding patient-years at risk pooled over 1993-1995. However, each SHR is determined by computing the expected number of first hospitalizations using the 1995 national first hospitalization rates. This represents a modest change over the 1996 ADR, and is due solely to the change in methodology used to obtain the national ESRD hospitalization rates. Also, except where indicated, units with fewer than 20 admissions in the three year period are excluded from the analyses to further ensure the stability of the calculated rates. Analyses are performed by census region as well as at the dialysis unit level based on unit-specific SHRs. Where appropriate, comparisons to the SMR are also made. ------------------------------------------------------------------------ Figure IX-7 [Image] Figure IX-7. Distribution of standardized first hospital admissions ratio for each dialysis unit size. Ratios are standardized by sex, race, age and diabetic status. Facility sizes are determined by the expected number of first admissions for that unit in a three year period, 1993-1995. Those with fewer than twenty first admissions in that time are in the smallest group; those with twenty to fifty are in the next group; and those units with more than fifty first admissions are included in the last group. Ratios are given for the 10th percentile, the 25th percentile, the median, the 75th percentile and the 90th percentile. National average is 1.05. Source: Special Analysis. ------------------------------------------------------------------------ Figure IX-5 summarizes the distribution of SHRs and SMRs for dialysis units within each of the nine census regions shown in Figure IX-6. Smaller units have been excluded from these analyses, so the unit-specific ratios are reasonably stable. There is evidently some variability in the rates across region; however, this is to be expected as both the number of facilities and their sizes vary significantly within and across regions. The median SHR within each region is given by the solid line in each boxplot as well as on the map in Figure IX-6, and provides a snapshot of the typical dialysis unit in each region. For six of the nine regions, the median SHR exceeded 1.0, indicating hospitalization rates above the ESRD norm; these regions are shaded in Figure IX-6. The pattern observed here reflects national practice patterns of patient hospitalization; it is well known that hospitalization rates generally tend to be higher east of the Mississippi River. The national median SHR is 1.06, indicating somewhat less than 50 percent of the nations dialysis facilities are close to or below the national ESRD first hospitalization rate. Regions having the lowest SHRs are the Pacific, Mountain, and West North Central Regions. These trends are consistent with trends observed in studies of non-ESRD patients (Gornick, 1982) as well as past studies of hospitalization among ESRD patients (USRDS ADR, 1991, 1995, 1996). The highest rate of hospitalization occurs in the Northeast, followed closely by the East South Central, Middle Atlantic, and West South Central regions. These patterns are very similar to those reported in the 1995 ADR, where the rates are based on total admissions, and also the analyses reported in the 1996 ADR, which are based on the SHR. There is moderate positive correlation between these rates and the corresponding SMR. For example, five of the six regions having the highest SHRs also have the highest median SMRs. It is unclear which factors are primarily responsible for the geographic variation observed here; several major patient mix characteristics have been adjusted for, but other factors which have not been adjusted for may also be important. Analyses by Facility Size The SHRs can also be compared at the dialysis unit level. Figure IX-7 contains box plots and summary statistics describing the distribution of SHRs by unit size, measured here in terms of the expected number of yearly first admissions between January 1, 1993, and December 31, 1995. The variability and skewness in the distribution of the SHR decreases quickly with the increase in expected first admissions. For example, the interquartile ranges (75th minus 25th percentiles) of the SHRs across the three categories are respectively 1.05, 0.48, and 0.34, indicating a substantial drop in variability between units having fewer than 20 expected first hospitalizations versus those with 20-50. The median SHR does not change dramatically with unit size, however, and both patterns are consistent with that reported in the 1996 ADR (Figure VIII-8). These results clearly indicate the need to account for standard error of the SHR when interpreting the value of the SHR computed for a particular facility. This is particularly evident for smaller facilities. ------------------------------------------------------------------------ Figure IX-8 [Image] Figure IX-8. Distribution of standardized first hospital admissions ratios by type of unit, 1993-1995. Units are divided into three types: free standing dialysis units operating for a profit, free standing dialysis units not operating for a profit, and hospitals. Ratios are standardized for age, race, sex and diabetic status. Units with fewer than twenty first admissions in the three year period are excluded. Ratios are given for the 10th percentile, the 25th percentile, the median, the 75th percentile and the 90th percentile. Source: Special Analysis. ------------------------------------------------------------------------ Although the reported trends are similar, a direct comparison with Figure VIII-8 in the 1996 ADR needs to be done very cautiously. SHRs computed for the 1996 ADR were based on national rates that represented a three year average, and consequently were standardized to a population which reflected a "true" rate which lay somewhere in between that for 1991-1993. The SHRs reported here are actually standardized to 1995 national hospitalization rates. Since hospitalization in general has been decreasing over time, the 1995 national rates are in a relative sense farther away from the "true" rates for 1993 and 1994 than were the averaged rates over 1991-1993 from each of the years 1991-1993. This is a mathematical tautology and consequently the current SHRs may be slightly inflated relative to those reported in the 1996 ADR. Comparisons by Profit Status and Facility Type In Figure IX-8, the SHR distributions, by whether a dialysis unit operates for profit (yes/no) and/or type of unit (hospital or freestanding), are shown in box plots. Standardized hospitalization ratios, when adjusted for age, race, sex and cause of ESRD, are typically higher in freestanding for-profit dialysis units (Median SHR = 1.08). Both hospitals (mostly not-for-profit) and not-for-profit freestanding units have slightly lower median SHRs, being at 1.02 and 0.99, respectively. The variability in the rates is approximately the same, although there is somewhat more variability in the freestanding for-profit facilities than in the other two categories. ------------------------------------------------------------------------ Table IX-2 [Image] ------------------------------------------------------------------------ The dialysis policy literature has for years debated whether hospital outpatient dialysis units have a "sicker" (more comorbid conditions and more severe conditions) patient population than do the freestanding units. If one takes the SHR as a measure of patient case-mix severity (beyond age, race, sex, and diabetes), these estimates suggest that the freestanding for-profit units have a higher severity mix than do the not-for-profit dialysis units. If, on the other hand, one takes the SHR as a measure of resources used to produce good patient health, then these data suggest that freestanding for-profit-units use a higher level of inpatient treatment resources than do the not-for-profit units. This is an intriguing observation since the outpatient units will generally receive no reimbursement for dialysis treatments provided to their patients as an inpatient. It might also be argued that dialysis units do not admit patients to the hospital; it is the physician, and their financial incentives must be considered as well. In most situations, physicians who see patients in an outpatient dialysis unit may also have an arrangement with a hospital to provide inpatient dialysis which would complicate the interpretation of these hospitalization statistics. Clearly much more information is needed to discern among these many hypotheses. The comparison of SHRs to SMRs is also interesting. In Table IX-2, a single pooled rate is calculated for each facility type. A similar calculation is done for the SMR. The pooled first hospitalization rates are nearly identical to the median facility SHR values. This is not terribly surprising due to the symmetry apparent in the distribution of SHRs by facility type in Figure IX-8. It is, however, very interesting to note that the freestanding-not-for-profit dialysis units have both a lower SHR and SMR than do either the freestanding for-profit units or the hospital units. This is similar to what was reported in the 1995 and 1996 ADRs. Interestingly, this trend cannot be attributed solely to the geographic locations of the freestanding/not-for-profit units since the Pacific, Mountain and West-North-Central regions, which have the lowest overall SHRs, comprise a relatively low percentage of the total number of those units (< 30 percent). Depending on how one interprets the SHR, the SMRs become interesting indicators of outcomes. If SHRs reflect patient severity (beyond age, sex, race and diagnosis), then freestanding units (both profits and not-for-profit) have outcomes (mortality) that are in line with the severity of the patients. Hospital units, under this interpretation, have higher mortality than their severity indicator (SHR) would suggest (although not by much). If, however, SHRs measure the mix of input resources used with patient severity having an assumed constant grouping average across all three groupings, then patients of hospital outpatient dialysis units use less inpatient resources with the same outcomes as do the patients of freestanding for profit dialysis units. Again, more information is required on patient comorbid status and other unmeasured indicators regarding the practice of nephrology. Time Trends in the SHR The unit specific SHR has a relatively stable value across years; this is not altogether unexpected due to the small decrease in total admissions for 1991-1995. Table IX-3 describes all pairwise rank correlations between the SHRs within a dialysis unit over the years 1991-1995. These correlations are all above the table diagonal; for comparison, a similar analysis is provided below the table diagonal for SMRs. For example, the value of 0.57 in the row labeled 1992 and column labeled 1993 is the correlation between SHRs within a facility between 1992 and 1993. The value of 0.27 in the row labeled 1993 and column labeled 1992 is the correlation between SMRs within a facility between 1992 and 1993. ------------------------------------------------------------------------ Table IX-3 [Image] ------------------------------------------------------------------------ Each facility-based SHR or SMR in this analysis is, for a given year, standardized to the 1995 national first hospitalization or mortality rate as appropriate. Facilities with less than 5 expected first admissions or deaths have been excluded. The within-facility SHR correlations for 1991 vs. 1992, 1992 vs. 1993, etc. are seen to be approximately the same, averaging 0.59. That is, if one were to regress the ranks of the facility SHRs in 1995 on those for 1994, the slope of the regression line would be positive with an R2 approximately equal to 0.35. The corresponding average SMR correlation is approximately 0.25, which is much lower. The average 2 year and 3 year facility SHR correlations are respectively about 0.5 and 0.45, while those for the SMR are about 0.26 and 0.21. These correlations indicate that a dialysis unit with a high SHR in one year is reasonably likely to have a high SHR over the next two to three years. This is in marked contrast to the correlation between the SMRs for those units in the same period of time, which are relatively low (maximum correlation = 0.28). The lower correlation seen in the SMR is partly due to the fact that it is based on fewer events (i.e. there are fewer deaths than hospitalizations) and is hence subject to greater variability. Consequently, the SMR for a dialysis in a single year does not necessarily predict their SMR in future years. Hospitalization has declined in recent years. In Figure IX-9, we present the yearly SHR for 1991-1995, standardized to 1995 national first hospitalization rates. Each is a single pooled adjusted rate calculated based on data from all dialysis units in a given year; a 95 percent confidence interval for the SHR’s in 1991-1994 is also given. The SHR for 1995 is by construction fixed at 1.0 and thus no confidence interval is given. Since each SHR is standardized to 1995 national rates, an SHR larger than one for, say, the year 1991 implies that the observed number of (first) hospitalizations in 1991 is higher than what is expected in 1995 for a population having the same at-risk time within each age/race/sex/diabetes group. In turn, this implies a drop in the hospitalization rate over 1991-1995. The results in Figure IX-9 indicate that dialysis patients are experiencing less hospitalization overall with each passing year, with about a 15 percent drop in (first) admissions between 1991 and 1995. The fact that the 95 percent confidence intervals do not overlap indicate that the SHR’s are statistically significantly different from one another. There is correlation between SHR’s in successive years that is not accounted for in this comparison; however, due to the large sample sizes involved, such correlation will not qualitatively alter the results observed here. Results given earlier (see Table IX-1) indicate that this drop in hospitalization is complemented by an approximately 17 percent reduction in the average number of days spent in the hospital each year. ------------------------------------------------------------------------ Figure IX-9 [Image] Figure IX-9. Standardized first hospital admissions ratio for all dialysis units over time, 1991-1995. Ratios are standardized for age, race, sex and diabetic status. Ratios are calculated by summing all observed first hospitalizations for each year and dividing them by the respective sum of expected first admissions. Ninety-five percent confidence intervals are indicated (as +/- value) for 1991-1994; but not for 1995 since rates are standardized to 1995. Source: Special Analysis. ------------------------------------------------------------------------ These reported decreases in hospitalization over the 1991-95 period are very dramatic. Chapter X of this report provides some estimates of the spending for inpatient care for both dialysis and transplant patients. The 1995 experience, compared to the average of the 1991-94 period, Medicare spending for inpatient care increased by 6 percent per patient year at risk. If the two indicators of hospitalization (SHRs reported in this Chapter and the spending per patient year at risk for inpatient care reported in Chapter X) are consistent then the spending per hospitalization episode must be increasing. Such phenomena would be consistent with a process of more severe cases being treated as an inpatient as less severe cases are shifted to the outpatient setting. Analyses of this topic are worthy of additional research that are beyond the scope of this current report. Nonetheless the substantial decreases in rates of hospitalization reported above are most intriguing and should be considered good news. But until these findings are repeated with either independent data from outside the Medicare system, or with independent data within the Medicare system (e.g. physician payments), then caution in the interpretation of these trends in SHRs would be prudent. References: End-Stage Renal Disease Research Report 1992. U.S. Department of Health and Human Services; Health Care Financing Administration; Bureau of Data Management and Strategy. September 1994. Gornick M. Trends and regional variations in hospital use under Medicare. in: Rothberg DL (ed). Heath and Co., Lexington MA. 1982, pp. 131-184. Habach G, Port FK, Mauger E, Wolfe RA, Bloembergen WE. Hospitalization among US dialysis patients: Hemodialysis versus peritoneal dialysis. J Am Soc Nephrol 1995, 5:1940-1948. Hoem J. Statistical analysis of a multiplicative model and its application to the standardization of vital rates: A review. Int Statist Rev 1987, 55:1191-52. Lehmann E. Nonparametrics: Statistical Methods Based on Ranks. San Francisco: Holden-Day. 1975. Port FK, Held PJ, Nolph KD, Turenne MN, Wolfe RA. Risk of peritonitis and technique failure by CAPD connection technique: a national study. Kidney Int 1992, 42: 967-974. Ross, S. Stochastic Processes. New York. Wiley. 1983. Strawderman RL, Levine G, Hirth RA, Port FK, Held, PJ. Using USRDS generated hospitalization tables to compare facility-specific ESRD hospitalization rates to national rates. Kidney Int 1996, 50: 571-578. United States Renal Data System. USRDS 1991 Annual Data Report. National Institutes of Health, National Institutes of Diabetes and Digestive and Kidney Diseases. Bethesda, MD, 1991. United States Renal Data System. USRDS 1995 Annual Data Report. National Institutes of Health, National Institutes of Diabetes and Digestive and Kidney Diseases. Bethesda, MD, 1995. Wolfe RA. The standardized mortality ratio revisited: Improvements, innovations, and limitations. Am J Kidney Dis 1994; 24:290-297. Wolfe RA, Gaylin DS, Port FK, Held PJ, Wood CL. Using USRDS generated mortality tables to compare local ESRD mortality rates to national rates. Kidney Int 1992; 42:991-996.