Avorn J, Bohn RL, Levy E, Levin R, Owen WF, Winkelmayer WC, Glynn RJ. Nephrologist Care and Mortality in Patients With Chronic Renal Insufficiency. Arch Intern Med. 2002;162(17):2002–2006. doi:10.1001/archinte.162.17.2002
For patients with chronic renal insufficiency, rates of referral to nephrologists are highly variable, and little is known about the effect of such consultation on clinical outcomes. We sought to determine whether early or frequent access to nephrologist care prior to the initiation of dialysis was associated with a difference in mortality rates in the first year after dialysis began.
We identified all patients in the New Jersey Medicaid and Medicare programs who began maintenance dialysis during a 6-year period and who had been diagnosed with renal disease more than 12 months prior to dialysis. Use of nephrologist services was documented during this 1-year period, along with other clinical and sociodemographic variables. The outcome measure of our analysis was mortality in the first year after initiation of dialysis.
From multivariate analyses, we found that patients who did not see a nephrologist until 90 days or less before initiation of dialysis had a 37% higher likelihood of death in the first year of dialysis compared with patients with earlier referral (95% confidence interval, 1.22-1.52; P<.001). Similarly, those who saw a nephrologist on fewer than 5 occasions in the year prior to dialysis had a 15% higher mortality rate in the first year of dialysis compared with those who had had 5 or more nephrologist visits (95% confidence interval, 1.03-1.28; P = .01).
For patients with long-standing renal disease, earlier consultation with a nephrologist and more frequent specialist encounters is associated with lower mortality in the first year of dialysis. These findings need to be confirmed in younger and less indigent patients as well.
WITH INCREASING pressures on all aspects of health care expenditures, and growing concern about the cost-effectiveness of medical interventions, attention has turned to the role of the specialist in various domains of patient care. Many payers and health care systems have placed renewed emphasis on the role of the primary care physician as both coordinator of services and as "gatekeeper" in rationing access to costly interventions. When done optimally, this can reduce unnecessary expenditures and improve the coordination of care. However, excessive application of such pressures can result in denial of access to more costly services, even when these are clinically appropriate and may be cost-effective.
Chronic renal disease is one arena in which these issues have been debated in recent years. End-stage renal disease (ESRD) is the only clinical condition for which the development of a given diagnosis automatically results in coverage of health care expenditures by the Medicare program, regardless of the age of the patient. In 1998, the total cost of care for the numerically small ESRD program was approximately $12 billion, rendering it the most costly single Medicare program.1 Since the late 1960s, the care of ESRD patients has been a lightning rod for societal questions about health care access, equity, and cost.
While dialysis care of ESRD patients is provided almost uniformly by nephrologists, the predialysis management of chronic renal insufficiency is a domain in which issues of specialist vs generalist care loom large. Little is known about the appropriate place of nephrologist care for patients with chronic renal insufficiency not requiring renal replacement therapy, a phase that often lasts for several years. Although extensive data are available through the Medicare ESRD program once patients receive chronic renal replacement therapy (ie, maintenance hemodialysis, peritoneal dialysis, or renal transplantation), it is much more difficult to assemble data during the period prior to their entry into this program. However, this is precisely the period during which it is most critical to study the effect of care on the subsequent clinical course of such patients.
Some have argued that early involvement of a nephrologist in such care can result in better clinical status at the time of initiation of dialysis, which in turn could result in an improved clinical outcome once dialysis has begun. Such preparation can include the timely and proactive development of vascular access required for hemodialysis, and optimization of hematologic, endocrine, nutritional, metabolic, and hemodynamic function in the face of progressive renal failure. In 1997, 55% of patients with chronic renal insufficiency were first seen only within 12 months prior to initiating dialysis, and 33% were first seen within 3 months of initiating dialysis.2,3 Several previous studies have demonstrated substantial heterogeneity in the management of patients with chronic renal insufficiency, perhaps related to the variability of timing and intensity of the interaction with nephrologists. The annual mortality rate for ESRD patients is approximately 20% per year, and half these deaths are attributed to cardiovascular complications.4 This mortality translates into a life expectancy that is only 16% to 37% that of the age-, sex-, and race-matched general population.2 The highest death risk occurs during the incident year of renal replacement therapy, and is subsequently lower.1,5
The effect of delayed referral to a nephrologist on morbidity and mortality has been examined in a limited manner.6 In a small study,7 patients receiving care from a nephrologist had shorter hospitalizations to initiate hemodialysis than did patients cared for by nonnephrologists, or those who received no medical care at all. Similarly, Hakim et al8 reported that hospital stays were longer and more costly for patients with chronic renal insufficiency referred to a nephrologist relatively late in the course of their disease. Other studies have examined the association of late nephrologist referral and mortality in new ESRD patients, but have produced conflicting results.9- 12 These studies used different definitions of late vs early referral, and were limited to small and highly selected samples.
The present study is the most comprehensive to date to examine this question in a very large population of patients cared for in typical settings. It was designed to examine the association between the utilization of nephrologist services by patients with chronic renal insufficiency during the year before the initiation of chronic dialysis, and mortality during the first year of renal replacement therapy.
We identified all patients who began maintenance dialysis between 1991 and mid-1996, and had been active participants in either the Medicare or Medicaid programs of the state of New Jersey for at least the prior 12 months. All personal identifiers were transformed into coded study numbers to protect confidentiality. Maintenance dialysis was identified by the International Classification of Diseases, Ninth Revision (ICD-9) and Current Procedural Terminology codes for hemodialysis, peritoneal dialysis, renal transplantation, and ESRD. The first record of maintenance dialysis during this period was referred to as the index claim. To ensure eligibility in the Medicare or Medicaid programs, patients were required to have had at least 1 medical encounter of any kind in the year prior to their first maintenance dialysis procedure. In addition, we required that the first diagnosis of renal disease have occurred more than 1 year prior to the initiation of dialysis, to exclude patients with new-onset renal insufficiency who may not have had the opportunity for many encounters with a nephrologist prior to dialysis because of the time course of their condition.
Patients were excluded if they had no second dialysis following the index procedure but survived more than 1 month, if they had more than 2 months between 2 consecutive dialysis procedures, or if their health care providers could not be identified. All patient identifiers were transformed into anonymized untraceable study numbers in all analyses to protect confidentiality.
For each patient, we then defined age, sex, and race on the date of initiation of maintenance dialysis. We also roughly characterized socioeconomic status at this point by defining whether the patient had been enrolled in Medicaid or the New Jersey Pharmaceutical Assistance to the Aged and Disabled (PAAD) program, a state-specific program that reimburses drug expenditures for state residents of modest income who are not indigent enough to qualify for Medicaid. We then defined the frequency and timing of visits with a nephrologist in the 12 months prior to initiation of maintenance dialysis. Physician specialty was identified by Medicare and Medicaid specialty codes as well as by Unique Physician Identification Numbers assigned to all practicing physicians. For each physician encounter, provider numbers were searched for the specialty code for nephrologist.
We also extracted information on all hospitalizations, physician visits, procedures, and nursing home care received by these patients during this period. This made it possible to identify all diagnoses assigned to these patients by all clinicians who cared for them, including specific renal diagnoses as well as comorbidities such as hypertension, diabetes mellitus, congestive heart failure, ischemic heart disease, and other relevant conditions (Table 1). Death rates were measured in the year after maintenance dialysis was begun, using data from Medicare and Medicaid enrollment files.
Referral patterns to a nephrologist were studied using 2 definitions. Late referral was defined as a patient having a first nephrologist encounter 90 days or less before the initiation of maintenance dialysis. A second definition, that of frequency of nephrologist care, was defined as the number of encounters with a nephrologist during the 12 months prior to initiation of maintenance dialysis. This variable was dichotomized: 5 or more visits vs less than 5 visits.
We developed a proportional hazards regression model to predict the likelihood of death in the 12 months following initiation of dialysis. Age, sex, race, PAAD/Medicaid membership as proxy for socioeconomic status, and underlying renal diagnosis (categorical) were forced into the model. Then we added the number of predialysis nephrologist visits and the variable for late referral, respectively, into 2 separate models. Finally, we tested all other covariates (categories of comorbidity, Charlson comorbidity score, physician characteristics) for individual significance or confounding. We also tested for significance of potential effect modifiers, most importantly the interaction between age and timing of referral.13
We identified 17 884 patients who underwent maintenance dialysis at some point during the period 1991 to 1996; of these patients, 12 557 had adequate baseline data for a full year prior to dialysis in Medicaid and/or Medicare to permit further study. In this population, 5242 patients had their first renal diagnosis at least 1 year prior to the initiation of dialysis and were therefore eligible for further study. Most of the patients without a previous diagnosis of kidney disease who were eliminated in this step can be assumed to have undergone one-time or short-term renal replacement therapy as a consequence of acute renal failure. Six hundred twenty-six patients in this group of 5242 patients had less than 30 days of renal replacement therapy, indicating that they had acute renal failure. Five hundred ninety-nine patients had more than 2 months without claims for renal replacement therapy and survived without additional dialysis care, and 1003 lacked adequate data describing their health care providers. This left a study population of 3014 patients. Of these, 1430 (47.4%) died in the first year of dialysis.
As seen in Table 1, the study population included large numbers of older patients, with 1288 (42.7%) between the ages of 65 and 74 years and 1063 (35.3%) aged 75 to 84 years. There were slightly more men than women (56.2% vs 43.8%), and 25.7% were nonwhite. The most common renal diagnoses specified were hypertensive nephropathy and diabetic nephropathy. Approximately one third of patients (34.5%) did not see a nephrologist until 90 days or less before their first renal replacement therapy. Half of the patients (50.5%) had fewer than 5 nephrologist consultations in the year prior to renal replacement therapy.
As expected, age was a strong predictor of the likelihood of mortality, with the risk ratio for death (compared with a referent group aged <45 years) increasing systematically from 2.97 for patients aged 45 to 54 years, to 11.53 for patients 85 years and older (Table 2). After controlling for clinical diagnoses (see below), we found no association between sex and mortality. However, blacks were found to have a lower risk of mortality, as reported in other studies of ESRD,14 with a risk ratio of 0.77 (P<.001). Individuals of race other than white or black were at even lower risk of death in their first year on renal replacement therapy (risk ratio, 0.65; P = .002). Being enrolled in either the New Jersey Medicaid or PAAD programs, which served as a proxy for lower socioeconomic status, was not significantly predictive of higher mortality (risk ratio, 0.97; P = .57).
After controlling for age, race, socioeconomic status, and the presence of all renal diagnoses recorded in the year before dialysis (Table 2), we examined the risk of death independently associated with late referral to a nephrologist as defined above, using proportional hazards regression analysis. Patients with late referral (first nephrologist consultation ≤90 days before initiation of dialysis) had a 37% increase in risk of death in the first year of dialysis compared with patients whose first nephrologist encounter occurred more than 90 days before the start of dialysis (95% confidence interval, 22%-52%; P<.001).
We next replaced this variable in the model with a variable defining the frequency of nephrologist visits during the 12 months prior to initiation of dialysis. Here again, after controlling for all other variables studied, patients who had seen a nephrologist fewer than 5 times in the year prior to initiation of dialysis had a 15% higher risk of death in the subsequent year, compared with those who saw a nephrologist more frequently during that period (P = .01). Adjusting for the Charlson comorbidity score and for other specific nonrenal clinical conditions (eg, diabetes, ischemic heart disease) yielded essentially identical results (risk ratiolate referral, 1.35; P<.001). In using interaction terms to study whether the effect of late referral differed by age group, we found that the effect was not significantly changed by the age of the patient.
These findings are based on what is, to our knowledge, the largest study to date of predialysis patients, and raise important questions about access to and utilization of nephrologist services in this vulnerable population. The excess mortality seen in patients with chronic renal insufficiency who have late or infrequent consultation with nephrologists does not appear to be an artifact caused by differences in patients' ages, sex, cause of their renal disease, socioeconomic status, or other comorbid conditions.
The present study may help reconcile the conflicting results among previous, smaller studies. While Ifudu et al9 (United States), Innes et al10 (United Kingdom), and Sesso et al11 (Brazil) had found an increase in mortality in patients with delayed nephrological care, Roubicek et al12 (France) did not find an increase in 1-year or 5-year mortality in patients who first saw a nephrologist less than 4 months prior to onset of dialysis. However, differences in methodology and populations under study probably confounded these previous results.15 The largest population in which this question was studied previously contained only 270 subjects,12 compared with the 3014 in the present report. The failure to find a difference in survival between patients referred to a nephrologist late vs early in previous studies is probably attributable to selection bias, or problems in generalizability from one population to another, particularly when cross-national health system issues are considered. Friedman15 has pointed out that the frail or elderly patients most vulnerable to delayed referral may be less likely to receive renal replacement therapy in France, while such patients with high comorbid health status and age, who may benefit most from timely nephrologist consultation, are likelier to receive treatment in the United States.
Although extensive outcome data are available through Medicare and its associated ESRD registries, it is more difficult to assemble data describing patients prior to their renal replacement therapy. However, this period is critical to study the effect of care on the subsequent clinical course of such patients. In evaluating these findings, some limitations of the study's reliance on Medicare and Medicaid data must be kept in mind. For patients younger than 65 years, the only source of predialysis information was care received through the Medicaid program, raising concerns about whether data from this group of patients can be generalized to younger, nonindigent patients with these conditions. However, for all patients who began dialysis after age 66, predialysis enrollment in Medicare was virtually universal, and we found no difference in the effect of nephrologist referral in this age group compared with younger patients (P = .21). Second, because the diagnoses were based in part on data from inpatient and outpatient encounter claims from the Medicare and Medicaid programs, there is a possibility of misclassification with regard to the cause of ESRD or other clinical conditions. Assessment of renal diagnosis was approached differently in the present study compared with United States Renal Data System (USRDS) data. In the USRDS data set, renal diagnosis is drawn from the HCFA [Health Care Financing Administration] 2728 form, where providers of renal replacement therapy enter the most likely renal diagnosis for a given patient. Instead, we screened for all renal diagnoses assigned in the year prior to onset of maintenance dialysis by all physicians and hospitals caring for the study subjects. Hence, renal diagnoses are not mutually exclusive in our study. Nonetheless, even in studies using primary medical records, the attribution of ESRD to a particular clinical cause is often difficult and often suspect, because most patients with chronic renal insufficiency do not undergo diagnostic renal biopsy; it is not unusual for patients to receive the default diagnosis of "hypertensive nephropathy" or "diabetic nephropathy" in such primary sources.
We also considered possible confounding by nonrenal clinical conditions, such as diabetes, hypertension, malignancy, congestive heart failure, or coronary heart disease. As expected, there was high collinearity between some comorbidities (eg, diabetes) and their related renal diagnoses (eg, diabetic nephropathy). When adding comorbid conditions to the model, or when replacing renal comorbidities with nonrenal comorbid conditions, the observed findings were virtually identical (risk ratiolate referral, 1.35; P<.001). It is possible that the sudden onset of rapidly progressive renal failure might have been associated both with a higher death rate on dialysis as well as with a lower rate of nephrologist visits in the preceding year. However, this possibility was addressed by requiring that all patients studied have a diagnosis of chronic renal failure extending at least a full year prior to the initiation of dialysis.
Our findings cannot at this point resolve the question of whether the utilization of nephrologist services may also serve as a marker for other characteristics that may also be important risk factors for mortality in ESRD. Socioeconomic status does not appear to play a role, since we did not find an increased adjusted risk of death in Medicaid or PAAD patients compared with those who were less indigent. However, use of specialist services may correlate with other issues of access and quality of care, which in themselves may play an important role in outcomes. For example, patients who are referred to a nephrologist may have a primary care physician who is also more conscientious at managing other aspects of their care, such as hypertension or nutrition. Or, such patients may themselves be more compliant with recommendations for their care, or more adept at navigating the health care system.
Alternatively, the nephrologist involvement may make referred patients more likely to be treated with erythropoietin for anemia, possibly lowering the risk of death, or to have a permanent vascular access created for maintenance dialysis, especially a primary fistula. This in turn may reduce infectious risk during renal replacement therapy and/or improve dialysis doses. Other testable hypotheses are that patients who see a nephrologist earlier and more often have better treatment of comorbid conditions associated with progressive renal failure, such as hypertension, malnutrition, and/or hyperphosphatemia, or may be better prepared psychologically for dialysis, resulting in better compliance with dietary and/or fluid regimens. If processes of care for pre-ESRD patients differ across specialties and these differences can have a beneficial impact on ESRD mortality, follow-up studies will be necessary to further define these and other possibilities. This may be of particular importance if manpower constraints in the coming years limit the number of nephrologists available to provide such additional care; might there be some particular interventions that could also be performed by a primary care physician or another consultant if nephrologist care were in short supply?
If these findings are replicated in other settings, especially in younger and less indigent cohorts, it will be important to further define the mechanism through which early and/or more frequent nephrologist input appears to have a beneficial effect on patient outcomes. If confirmed, such findings could have important implications for quality improvement programs,16,17 manpower projections,18,19 the care of particular high-risk populations,20 as well as health policy and resource allocation decisions.21
Accepted for publication February 6, 2002.
This work was supported by grant R0-1-HS09398 from the Agency for Healthcare Research and Quality. Additional support was provided by the Health Care Financing Administration.