Abstract
Abstract. The association between hematocrit level and future hospitalization risks in hemodialysis patients has not been fully investigated on a national level. A total of 71,717 prevalent Medicare hemodialysis patients who survived a 6-mo entry period from July 1 through December 31, 1993 were studied, and their risk of hospitalizations was evaluated the next year. Five hematocrit groups were defined from Medicare recombinant human erythropoietin-treated patients: <27%, 27 to <30%, 30 to <33%, 33 to <36%, and ≥36%. A Cox regression model was used to investigate the association between hematocrit level and the risk of first hospitalization, and the Andersen-Gill regression model evaluated multiple hospitalizations during the next year, adjusting for patient comorbidity and severity of disease. Compared with the baseline group of 30 to <33%, patients with hematocrit levels <30% had a 14 to 30% increased risk of hospitalization without disease severity adjustment (p = 0.0001) and a 7 to 18% increased risk with disease severity adjustment (p = 0.0001). Patients in the 33 to <36% group had the lowest risk at 0.93 and 0.88 (p = 0.0001), with and without adjustment for disease severity. It is concluded that patients with hematocrits of <30% have an increased risk of future hospitalization, with hematocrit levels between 33 and 36% having the lowest associated risks.
The treatment of anemia secondary to chronic renal failure has been altered dramatically since the clinical introduction of recombinant human erythropoietin (rhEPO) (1). For nearly 10 yr, the clinical use of rhEPO to enhance the erythropoietic response has focused on physiologic and functional outcomes, including exercise capacity, brain electrophysiology, cardiac function, and quality of life (2,3,4,5,6,7,8,9,10,11,12,13,14). Few data are available on a major morbidity parameter in hemodialysis patients, namely hospitalization. The evaluation of hematocrit as a predictive variable for future hospitalization is complicated by the lack of detailed data on comorbidity and severity of disease, both of which are known to influence hematocrit levels, rhEPO responsiveness, and hospitalization outcomes (1).
To study the possible associations of hematocrit level with future hospitalization, we analyzed the Health Care Financing Administration (HCFA) end-stage renal disease (ESRD) registration and claims data, assessing the complex interactions of patient comorbidity, severity of disease, and hematocrit level, determining their predictive value for future hospitalizations. This report summarizes the associations we observed.
Materials and Methods
Medicare Patients
All ESRD patients in the United States have been entitled to Medicare coverage since 1972. ESRD patients under the age of 65 yr have a 90-d waiting period before receiving Medicare coverage for in-center hemodialysis. All Medicare rhEPO claims are processed on detailed Uniform Bill forms (UB82/UB92) and require reporting of a hematocrit level prior to the last rhEPO dose of the billing period as a condition for reimbursement. Patients not receiving rhEPO do not have hematocrit levels documented in the Medicare system and are not included in this analysis. Medicare patients with rhEPO claims and with hematocrit levels are the source of data for our hospitalization risk model.
Study Design
We retrospectively evaluated prevalent Medicare hemodialysis patients surviving a 6-mo entry period from July 1 through December 31, 1993. Patient age, gender, race, and renal diagnosis data were obtained from the HCFA 2728 Medical Evidence forms. The 6-mo entry period was selected in order to adequately assign patients within hematocrit groups based on the hematocrit data provided on the rhEPO claims as well as to assess patient severity of disease (see below). The follow-up period was January 1 through December 31, 1994. All patients were on hemodialysis throughout the entry and follow-up periods.
Patient Characteristics
Patient characteristics included age at the beginning of the entry period, gender, race, and renal diagnosis (including diabetic status). Patient comorbidity was obtained from both Part A (1984-1993) and Part B (1991-1993) Medicare claims and profiled as described previously (15). Major medical conditions included atherosclerotic heart disease, congestive heart failure, peripheral vascular disease, cerebrovascular accident/transient ischemic attacks, other cardiac disease (valvular heart disease, arrhythmias, and pacemakers), cancer excluding skin malignancies (except for malignant melanoma), chronic obstructive pulmonary disease, liver disease, gallbladder disease, and gastrointestinal diseases associated with bleeding.
Severity of Disease (Factors Affecting Hematocrit Levels)
Medical conditions associated with blood loss may significantly affect hematocrit levels. Examples of these conditions include vascular access procedures and conditions requiring blood transfusions. Hospitalizations also affect hematocrit levels due to increased blood testing as well as acute medical complications that may create reduced bone marrow responsiveness to rhEPO, which reduces hematocrit levels during the hospitalization (2). These severity of disease measures were assessed for their relationship to the entry period hematocrit level and their predictive value on future hospitalization risks.
Hematocrit Levels
Medicare reimbursement for rhEPO administration requires documentation of the last hematocrit level before the last rhEPO dose given during the billing period. To define a consistent hematocrit level, only those patients with at least four rhEPO claims in the 6-mo entry period were included in the study. At least four rhEPO claims were needed to ensure a minimum of 3 mo of hematocrit data, with a mean of 5.1 mo of the 6-mo entry period. A mean hematocrit was calculated for each patient with subsequent hematocrit groupings of <27% (hemoglobin <9.0 g/dl), 27 to <30% (hemoglobin 9.0 to <10.0 g/dl), 30 to <33% (hemoglobin 10.0 to <11.0 g/dl), 33 to <36% (hemoglobin 11.0 to <12.0 gm/dl), and ≥36% (hemoglobin ≥12.0 g/dl).
Study End Points
Time to first hospitalization and cause-specific hospitalization was assessed, as well as multiple hospitalizations during the follow-up period. Cause-specific hospitalizations were evaluated by partitioning the International Classification of Diseases (ICD-9) codes for principle diagnoses into infectious (all types from all organ systems), cardiovascular, and other groupings. Patients were censored at death, modality change (hemodialysis to peritoneal dialysis or to transplantation), or at the end of the follow-up period.
Statistical Analyses
Explanatory Variables. Explanatory variables included age, race, gender, comorbidity, prior ESRD time, number of vascular access procedures, number of blood transfusions, and entry period hospital length of stay in days. Hematocrit groupings were evaluated after adjustments for the prior variables. χ2 was used to compare the differences in hospitalizations by hematocrit group in the univariate analysis.
Cox and Andersen-Gill Regression. The time from January 1, 1994 to the first hospitalization was used in the Cox proportional hazards model to assess the relative risk (16). An Andersen-Gill multiple event model (17), a modification of the Cox regression, was used to evaluate outpatient time between each hospitalization event as a risk ratio. By using multiple hospitalizations, the number of events is increased three- to fourfold, compared to single hospitalizations. All models were stratified by diabetic status to address the proportional hazards assumption (18). Patient outcomes during the follow-up period were adjusted for prior ESRD time, since patients with shorter time on ESRD have higher death rates compared to patients with longer time on dialysis (19). Comparison of results with and without severity of disease were evaluated to address the potential interaction between sicker patients and lower hematocrit levels. The reference patients were male and white, with no comorbidity, no access procedures or blood transfusions, and no hospitalizations during the entry period. The reference hematocrit group was 30 to <33%. All probability values were two-tailed.
Sensitivity Study. A sensitivity analysis was performed to assess the robustness of the results with respect to the number of patients in the 33 to <36% hematocrit group. We randomly selected samples of 50, 40, 30, 20, and 10% of the patients, using the last two digits of the Social Security number (20,21), rerunning the Cox regression model for each sample to determine the point at which significance of the findings is lost.
Results
Patient Characteristics
The 1993 study cohort contained 71,717 patients, with a mean age of 60.2 yr, 49% white, 41% black, 52% female, and 31% diabetic, from the primary renal diagnosis. This was comparable to the United States in-center hemodialysis population at a mean age 59.0 yr, 55% white, 39% black, 49% female, and 33% diabetic (see Tables C4 and C5 of the 1997 USRDS Annual Data Report) (22). The study cohort represented 78% of the patients surviving the 6-mo entry period and are more completely described in our mortality study of the same population (23).
Comorbidity, Disease Severity, and Hematocrit Group in the Entry Period
We analyzed the relationship between the number of preexisting comorbid conditions, the hematocrit level during the entry period, and the total hospital days for diabetic and nondiabetic patients (Table 1). Overall, diabetic patients had 62% more hospital days during the entry period compared with nondiabetic patients. There was a consistent pattern of increased hospital days in the entry period as the number of comorbid conditions increased and the hematocrit decreased.
Hospital days during the entry period for diabetic patients by number of comorbidities and hematocrit (Hct) levela
Hospitalizations in the Follow-Up Period
The relationship between hematocrit level and future hospitalization days per patient year is shown in Figure 1. The rapid falloff in patients at the higher levels is consistent with the restrictive Medicare reimbursement policies for patients with hematocrits >36%. The future hospital days per patient year monotonically decreases as the hematocrits rise to 36%. Hematocrits >36% were associated with more variable hospital days per patient year. The hospital days by hematocrit group is shown in Table 2. There was a steady decrease in admission rates and hospital days as patient hematocrits rose to the 33 to <36% range.
- Download figure
- Open in new tab
- Download powerpoint
Follow-up total hospital days per patient year and patient distribution by hematocrit level (unadjusted).
Unadjusted total hospital days and admission rates (per patient year) in the follow-up period by hematocrit group (all patients)
All-Cause Hospitalization Risk
We analyzed the adjusted risk of all-cause hospitalization in the follow-up period using the Cox proportional hazards model for the first hospitalization and the Andersen-Gill multiple event model for multiple hospitalizations (Tables 3 and 4). These analyses were performed with and without adjustments for severity of disease measures, which included blood transfusions, vascular access procedures, and total hospital days during the entry period. Compared with the baseline group of 30 to <33%, patients with hematocrit levels <30% had a 14 to 30% increased risk of future hospitalization without adjusting for disease severity, and a 7 to 18% increased risk with that adjustment. Patients with hematocrit levels in the 33 to <36% range had the lowest risk of future hospitalizations, at 0.93 and 0.88 (p = 0.0001), both with and without adjustment for disease severity, respectively. The severity of disease adjustment changed the magnitude, but not the direction, of the risks, except in the hematocrit group of ≥36. The latter variable results, both with and without severity of disease measures, may reflect the low sample size of 687 patients.
Cox proportional hazards model, stratifying on diabetes for the first hospitalization, without disease severity adjustmenta
Comparison of Cox proportional and Andersen-Gill multiplicative hazards model results for first hospitalization risk by hematocrit levels with and without severity of disease adjustmenta
The impact of disease severity on future hospitalization risk was highly significant (Table 4). Each unit of blood transfused in the entry period was associated with a future hospitalization risk of 1.05. A similar but smaller risk (1.03) was noted for each vascular access procedure performed during the entry period. When assessing the future risk of hospitalization, the impact of prior hospital days during the entry period was the largest among all of the factors adjusted for in the model. Therefore, prior hospital days are strong predictors of future risk of hospitalization.
Risk for Cause-Specific First Hospitalization
Cause-specific hospitalizations are shown in Table 5. Cardiovascular hospitalizations consisted of congestive heart failure, fluid overload, and cardiomyopathy (38%); ischemic heart disease (20%); circulatory system disorders (17%); cerebrovascular disease (11%); and other (14%). The infectious hospitalizations consisted of respiratory infection (37%), bacteremia/septicemia/viremia (20%), vascular access (18%), and other (25%). The differential effects relative to the specific hospitalization categories appear to be reflected in most of the risk factors. For example, females appear to have an 11% higher risk of infectious hospitalization as their male counterparts. Blacks and other minorities appear to have a lower hospitalization risk for all-cause, cardiac, and infectious hospitalization compared with whites. Individual comorbidity risks appear to have different associations based on the type of hospitalization. Atherosclerotic heart disease was associated with a 1.49 risk (95% confidence interval [CI], 1.43 to 1.56) for cardiovascular hospitalizations but only a 1.07 risk (95% CI, 1.02 to 1.13) for infectious hospitalizations. Chronic obstructive pulmonary disease was associated with a 1.48 risk of infectious hospitalization (95% CI, 1.41 to 1.56) compared to cardiovascular at 1.25 (95% CI, 1.19 to 1.31). Peripheral vascular disease was associated with an infectious risk of 1.31 (95% CI, 1.24 to 1.37) compared to cardiovascular at 1.18 (95% CI, 1.13 to 1.23). Congestive heart failure had a high cardiovascular hospitalization risk at 1.62, and infectious hospitalization risk was also high at 1.44.
The impact of patient characteristics and hematocrit levels on cause-specific first hospitalization risk without adjusting for severity of disease measuresa
The cause-specific risk by hematocrit group also shows a differential pattern. Patients in the <27% hematocrit group had a 1.32 cardiovascular (95% CI, 1.25 to 1.41) and a 1.57 infectious (95% CI, 1.47 to 1.68) hospitalization risk compared to the reference group of 30 to <33%. Patients in the hematocrit group 33 to <36% had the lowest all-cause, cardiovascular, and infectious hospitalization risk.
Sensitivity Analysis
Using the Cox regression model, we performed a sensitivity analysis on the risk of first hospitalization to determine the sample size required to detect the impact of the hematocrit group 33 to <36%. Figure 2 shows that at least 3000 to 4000 patients were required to detect the lower risk of hospitalization when severity of disease was included, whereas approximately 2000 patients were needed if disease severity was excluded.
- Download figure
- Open in new tab
- Download powerpoint
The impact of sample size on the mortality risk in the 33 to <36% hematocrit group, both with and without adjustment for severity of disease (number of access procedures, blood transfusions, and hospital length of stay), defined in the entry period.
Discussion
Many clinical studies of dialysis patients have associated increased hematocrit levels with improved cognitive function, reduced left ventricular hypertrophy, increased exercise tolerance, and improved qualtiy of life (24). The hematocrit levels achieved in these studies, however, were usually in the 30 to 33% range, with no studies evaluating hospitalization risks at this or other hematocrit levels (25).
Our study included a large national sample of 71,717 patients, which was representative of the in-center hemodialysis population. We have demonstrated that hematocrit levels in the entry period are highly associated with prior comorbidity and hospital days in the same period. Hematocrit levels are also associated with future hospitalizations as are other entry period variables, thereby demonstrating the confounding relationship between hematocrit level, comorbidity, and severity of disease.
To address these confounding effects and to more clearly define the associations between hematocrit level and the risk of hospitalization, we performed a detailed comparison of results with and without severity of disease adjustments. We assessed the impact of sample size in the same manner. The comparisons between the Cox and Andersen-Gill models showed the important impact of disease severity on the absolute value, but not the direction, of future hospitalization risk. Hematocrit levels <30% had a 7 to 31% increased risk of all-cause hospitalization with and without severity adjustments, respectively, compared with the hematocrit group of 30 to <33%. Higher hematocrit levels of 33 to <36% were associated with reproducible lower hospitalization risks.
The cause-specific hospitalization risks provide additional insight into the clinical factors associated with morbidity. The differential risk of cardiovascular and infectious hospitalization risks is apparent, particularly for patients with lower hematocrit levels.
The high cardiovascular and infectious hospitalization risks associated with congestive heart failure are unique within the comorbidity variables. The high infectious hospitalization risk with gastrointestinal bleeding comorbidity may reflect the risk of bacteremia from the gastrointestinal tract during periods when the bowel is compromised. The high infectious hospitalization risk in chronic obstructive pulmonary disease is consistent with clinically compromised pulmonary function in these patients.
The association between lower hematocrit levels and higher infectious hospitalization risk is of concern. The lower hematocrit group <27% may reflect poor rhEPO response secondary to chronic inflammatory conditions (2), which may predispose patients to a higher risk of infectious hospitalization. This is suggested by the poor white cell function in patients with elevated serum ferritins, as noted by Patruta et al. (26). Reduced bacterial killing and oxidation activity was noted in dialysis patients with ferritins >650 μg/L, iron saturation levels <20%, and hematocrits <30%, compared to patients with normal iron indices. The alternative hypothesis of increased infectious risk in severely anemic patients is also possible. Unfortunately, these interrelated factors are not separable in this observational study.
Our sensitivity studies demonstrate the importance of sample size in detecting the hospitalization associations in the 33 to <36% hematocrit range. When disease severity is included in the analysis, it appears that at least 3000 to 4000 patients are required in the higher hematocrit group of 33 to <36% to detect a significantly lower associated hospital risk.
Several limitations are important to consider in assessing the results of our study. The lack of dialysis therapy data in this large population study is a reality of the data sources (Medicare claims) and the study's large sample size (approximately 72,000 patients). The renal network core indicator project is a possible source for this data, but its use was beyond the scope of our study (27). Other data sources, such as the USRDS case-mix adequacy study or the Dialysis Mortality and Morbidity Study, included patients from the same period as our study, but the limited sample size of 24,000 patients may not have the power required to detect the impact of higher hematocrit levels on the risk of hospitalization (22). The impact of dialysis therapy on hematocrit-associated mortality in the Madore et al. study showed little effect on the results beyond case-mix factors of age, gender, race, renal diagnosis, and albumin for hemoglobins >90 g/L (28). Albumin level did change the magnitude of the mortality results for hemoglobin levels <90 g/L but had no impact for higher hemoglobins. Although the Medicare data do not contain albumin levels, the impact of comorbidity and severity of disease on the magnitude of the results was clear at all hematocrit levels, which is a more consistent result than adjustment for albumin alone (28). The persistent impact of higher hematocrits after disease severity adjustment reduces the likelihood that patients with lower hematocrits have higher risk secondary only to more severe disease.
The association of the higher hematocrits with lower hospitalization rates does not imply causality. Other factors may influence patient responsiveness to rhEPO and subsequent hematocrit levels. We attempted to address some of these factors, but further study is necessary.
Conclusions
The introduction of epoetin alpha in 1989 to correct the anemia of chronic renal failure has led to improvements in hematocrit levels. To evaluate the morbidity impact of higher hematocrits, an entry period epidemiologic model was created to assess the hematocrit level, comorbidities, and severity of disease. We demonstrated the confounding effects of patient characteristics, comorbidity, and severity of disease parameters on hematocrit level and their association with future patient morbidity. After these adjustments, patients with hematocrits of 33 to <36% had the lowest associated risks of future hospitalization.
Acknowledgments
Acknowledgments
This work was funded in part by an unrestricted research grant from the Minneapolis Medical Research Foundation and Amgen, Inc. The authors thank the entire analytical and technical staff at Nephrology Analytical Services at the Minneapolis Medical Research Foundation/Hennepin County Medical Center for their enormous efforts in conducting this complex study. We also thank Tom Arnold and Mike Hadad at the Health Care Financing Administration for their technical assistance. The assistance of Dana Knopic and Susan Everson in preparation of the manuscript is greatly appreciated.
Footnotes
-
American Society of Nephrology
- © 1999 American Society of Nephrology