On a preference-based instrumental variable approach in reducing unmeasured confounding-by-indication.

Treatment preferences of groups (e.g., clinical centers) have often been proposed as instruments to control for unmeasured confounding-by-indication in instrumental variable (IV) analyses. However, formal evaluations of these group-preference-based instruments are lacking. Unique challenges include the following: (i) correlations between outcomes within groups; (ii) the multi-value nature of the instruments; (iii) unmeasured confounding occurring between and within groups. We introduce the framework of between-group and within-group confounding to assess assumptions required for the group-preference-based IV analyses. Our work illustrates that, when unmeasured confounding effects exist only within groups but not between groups, preference-based IVs can satisfy assumptions required for valid instruments. We then derive a closed-form expression of asymptotic bias of the two-stage generalized ordinary least squares estimator when the IVs are valid. Simulations demonstrate that the asymptotic bias formula approximates bias in finite samples quite well, particularly when the number of groups is moderate to large. The bias formula shows that when the cluster size is finite, the IV estimator is asymptotically biased; only when both the number of groups and cluster size go to infinity, the bias disappears. However, the IV estimator remains advantageous in reducing bias from confounding-by-indication. The bias assessment provides practical guidance for preference-based IV analyses. To increase their performance, one should adjust for as many measured confounders as possible, consider groups that have the most random variation in treatment assignment and increase cluster size. To minimize the likelihood for these IVs to be invalid, one should minimize unmeasured between-group confounding.

Spectrum of cancer risk among US solid organ transplant recipients.

CONTEXT: Solid organ transplant recipients have elevated cancer risk due to immunosuppression and oncogenic viral infections. Because most prior research has concerned kidney recipients, large studies that include recipients of differing organs can inform cancer etiology. OBJECTIVE: To describe the overall pattern of cancer following solid organ transplantation. DESIGN, SETTING, AND PARTICIPANTS: Cohort study using linked data on solid organ transplant recipients from the US Scientific Registry of Transplant Recipients (1987-2008) and 13 state and regional cancer registries. MAIN OUTCOME MEASURES: Standardized incidence ratios (SIRs) and excess absolute risks (EARs) assessing relative and absolute cancer risk in transplant recipients compared with the general population. RESULTS: The registry linkages yielded data on 175,732 solid organ transplants (58.4% for kidney, 21.6% for liver, 10.0% for heart, and 4.0% for lung). The overall cancer risk was elevated with 10,656 cases and an incidence of 1375 per 100,000 person-years (SIR, 2.10 [95% CI, 2.06-2.14]; EAR, 719.3 [95% CI, 693.3-745.6] per 100,000 person-years). Risk was increased for 32 different malignancies, some related to known infections (eg, anal cancer, Kaposi sarcoma) and others unrelated (eg, melanoma, thyroid and lip cancers). The most common malignancies with elevated risk were non-Hodgkin lymphoma (n = 1504; incidence: 194.0 per 100,000 person-years; SIR, 7.54 [95% CI, 7.17-7.93]; EAR, 168.3 [95% CI, 158.6-178.4] per 100,000 person-years) and cancers of the lung (n = 1344; incidence: 173.4 per 100,000 person-years; SIR, 1.97 [95% CI, 1.86-2.08]; EAR, 85.3 [95% CI, 76.2-94.8] per 100,000 person-years), liver (n = 930; incidence: 120.0 per 100,000 person-years; SIR, 11.56 [95% CI, 10.83-12.33]; EAR, 109.6 [95% CI, 102.0-117.6] per 100,000 person-years), and kidney (n = 752; incidence: 97.0 per 100,000 person-years; SIR, 4.65 [95% CI, 4.32-4.99]; EAR, 76.1 [95% CI, 69.3-83.3] per 100,000 person-years). Lung cancer risk was most elevated in lung recipients (SIR, 6.13 [95% CI, 5.18-7.21]) but also increased among other recipients (kidney: SIR, 1.46 [95% CI, 1.34-1.59]; liver: SIR, 1.95 [95% CI, 1.74-2.19]; and heart: SIR, 2.67 [95% CI, 2.40-2.95]). Liver cancer risk was elevated only among liver recipients (SIR, 43.83 [95% CI, 40.90-46.91]), who manifested exceptional risk in the first 6 months (SIR, 508.97 [95% CI, 474.16-545.66]) and a 2-fold excess risk for 10 to 15 years thereafter (SIR, 2.22 [95% CI, 1.57-3.04]). Among kidney recipients, kidney cancer risk was elevated (SIR, 6.66 [95% CI, 6.12-7.23]) and bimodal in onset time. Kidney cancer risk also was increased in liver recipients (SIR, 1.80 [95% CI, 1.40-2.29]) and heart recipients (SIR, 2.90 [95% CI, 2.32-3.59]). CONCLUSION: Compared with the general population, recipients of a kidney, liver, heart, or lung transplant have an increased risk for diverse infection-related and unrelated cancers.

Facility-level interpatient hemoglobin variability in hemodialysis centers participating in the Dialysis Outcomes and Practice Patterns Study (DOPPS): Associations with mortality, patient characteristics, and facility practices.

BACKGROUND: Hemodialysis patients with larger hemoglobin level fluctuations have higher mortality rates. We describe facility-level interpatient hemoglobin variability, its relation to patient mortality, and factors associated with facility-level hemoglobin variability or achieving hemoglobin levels of 10.5-12.0 g/dL. Facility-level hemoglobin variability may reflect within-patient hemoglobin variability and facility-level anemia-control practices. STUDY DESIGN: Prospective cohort study. SETTING & PARTICIPANTS: Data from the Dialysis Outcomes and Practice Patterns Study (DOPPS; 26,510 hemodialysis patients, 930 facilities, 12 countries, 1996-2008) and from the Centers for Medicare & Medicaid Services (CMS; 193,291 hemodialysis patients, 3,741 US facilities, 2002). PREDICTORS: Standard deviation (SD) in single-measurement hemoglobin levels in hemodialysis patients in facility cross-sections (facility-level hemoglobin SD); patient characteristics; facility practices. OUTCOMES: Patient-level mortality; additionally, facility practices correlated with facility-level hemoglobin SD or patient hemoglobin levels of 10.5-12.0 g/dL. RESULTS: Facility-level hemoglobin SD varied more than 5-fold across DOPPS facilities (range, 0.5-2.7 g/dL; mean, 1.3 g/dL) and by country (range, 1.1 in Japan-DOPPS [2005/2006] to 1.7 g/dL in Spain-DOPPS [1998/1999]), with substantial decreases seen in many countries from 1998 to 2007. Facility-level hemoglobin SD was related inversely to patient age, but was associated minimally with more than 30 other patient characteristics and facility mean hemoglobin levels. Several anemia management practices were associated strongly with facility-level hemoglobin SD and having a hemoglobin level of 10.5-12.0 g/dL. When examined in CMS data, facility-level hemoglobin SD was positively associated with within-patient hemoglobin SD during the prior 6 months. Patient mortality rates were higher with greater facility-level hemoglobin SD (DOPPS: HR, 1.08 per 0.5-g/dL greater facility-level hemoglobin SD [95% CI, 1.02-1.15; P = 0.006]; CMS: HR, 1.16 per 0.5-g/dL greater facility-level hemoglobin SD [95% CI, 1.11-1.21; P < 0. 001]). LIMITATIONS: Residual confounding. CONCLUSIONS: Facility-level hemoglobin SD was associated strongly and positively with patient mortality, not tightly linked to numerous patient characteristics, but related strongly to facility anemia management practices. Facility-level hemoglobin variability may be modifiable and its optimization may improve hemodialysis patient survival.

Hepatitis C is a risk factor for death after liver retransplantation.

Retransplantation for liver allograft failure associated with hepatitis C virus (HCV) has been increasing due to nearly universal posttransplant HCV recurrence and has been demonstrated to be associated with poor outcomes. We report on the risk factors for death after retransplantation among liver recipients with HCV. A retrospective cohort of liver transplant recipients who underwent retransplantation between January 1997 and December 2002 was identified in the Scientific Registry of Transplant Recipients database. Cox regression was used to assess the relative effect of HCV diagnosis on mortality risk after retransplantation and was adjusted for multiple covariates. Of 1,718 liver retransplantations during the study period, 464 (27%) were associated with a diagnosis of HCV infection. Based on Cox regression, retransplant recipients with HCV had a 30% higher covariate-adjusted mortality risk than those without HCV diagnosis (hazard ratio [HR], 1.30; 95% confidence interval [CI], 1.10-1.54; P = 0.002). Other covariates associated with significant relative risk of death after retransplantation included older recipient age, presence in an intensive care unit (ICU), serum creatinine, and donor age. Additional regression analysis revealed that the increase in mortality risk associated with HCV was concentrated between 3 and 24 months postretransplantation, among patients age 18 to 39 at retransplant, and in patients retransplanted during the years 2000 to 2002. In conclusion, HCV liver recipients account for a considerable proportion of all retransplantations performed. Surprisingly, younger age predicted a higher mortality for recipients with HCV undergoing liver retransplantation. This may reflect a willingness to retransplant younger patients with an increased severity of illness or a more virulent HCV infection in this population. Although HCV was predictive of an increased risk of death, consideration of other characteristics of HCV patients, including donor and recipient age and need for preoperative ICU care may identify those at significantly higher risk.

Predicted lifetimes for adult and pediatric split liver versus adult whole liver transplant recipients.

Split liver transplantation allows 2 recipients to receive transplants from one organ. Comparisons of predicted lifetimes for two alternatives (split liver for an adult and pediatric recipient vs. whole liver for an adult recipient) can help guide the use of donor livers. We analyzed mortality risk for 48,888 waitlisted candidates, 907 split and 21,913 whole deceased donor liver transplant recipients between January 1, 1995 and February 26, 2002. Cox regression models for pediatric and adult patients assessed average relative wait list and post-transplant death risks, for split liver recipients. Life years gained compared with remaining on the waiting list over a 2-year period were calculated. Seventy-six splits (152 recipients) and 24 re-transplants resulted from every 100 livers (13.1% [adult] and 18.0% [pediatric] 2-year re-transplant rates, respectively). Whole livers used for 93 adults also utilized 100 livers (re-transplant rate 7.0%). Eleven extra life years and 59 incremental recipients accrued from each 100 livers used for split compared with whole organ transplants. Split liver transplantation could provide enough organs to satisfy the entire current demand for pediatric donor livers in the United States, provide more aggregate years of life than whole organ transplants and result in larger numbers of recipients.

Shared frailty models for recurrent events and a terminal event.

There has been an increasing interest in the analysis of recurrent event data (Cook and Lawless, 2002, Statistical Methods in Medical Research 11, 141-166). In many situations, a terminating event such as death can happen during the follow-up period to preclude further occurrence of the recurrent events. Furthermore, the death time may be dependent on the recurrent event history. In this article we consider frailty proportional hazards models for the recurrent and terminal event processes. The dependence is modeled by conditioning on a shared frailty that is included in both hazard functions. Covariate effects can be taken into account in the model as well. Maximum likelihood estimation and inference are carried out through a Monte Carlo EM algorithm with Metropolis-Hastings sampler in the E-step. An analysis of hospitalization and death data for waitlisted dialysis patients is presented to illustrate the proposed methods. Methods to check the validity of the proposed model are also demonstrated. This model avoids the difficulties encountered in alternative approaches which attempt to specify a dependent joint distribution with marginal proportional hazards and yields an estimate of the degree of dependence.

Immunosuppression and the risk of post-transplant malignancy among cadaveric first kidney transplant recipients.

The success of renal transplantation may be counterbalanced by serious adverse medical events. The effect of immunosuppression on the incidence of de novo neoplasms among kidney recipients should be monitored continuously. Using data from the Scientific Registry of Transplant Recipients, we studied the association of induction therapy by immunosuppression with antilymphocyte antibodies, with the development of de novo neoplasms. The study population included more than 41 000 recipients who received a cadaveric first kidney transplant after December 31, 1995, and were followed through February 28, 2002. Using Cox regression models, we estimated time to development of two types of malignancy: de novo solid tumors and post-transplant lymphoproliferative disorder (PTLD). We made adjustments for several patient demographic factors and comorbidities. Induction therapy was significantly associated with a higher relative risk (RR) of PTLD (RR = 1.78, p < 0.001), but not with a greater likelihood of de novo tumors (RR = 1.07, p = 0.42). Treatment with maintenance tacrolimus vs. cyclosporine showed a significantly different RR of developing de novo tumors for recipients with induction than for those not receiving induction (p = 0.024). These new estimates of the magnitude of malignancy risk associated with induction therapy may be useful for clinical practice.

Mortality and hospitalization in haemodialysis patients in five European countries: results from the Dialysis Outcomes and Practice Patterns Study (DOPPS).

BACKGROUND: Mortality and hospitalization rates are reported for nationally representative random samples of haemodialysis patients treated at randomly selected dialysis facilities in five European countries participating in the Dialysis Outcomes and Practice Pattern Study (DOPPS) (France, Germany, Italy, Spain and the UK). RESULTS: In the UK, 28.1% of haemodialysis patients received prior peritoneal dialysis treatment compared with 4.2-8.3% in other countries. Kidney transplantation rates ranged from 3.3 (per 100 patient years) in Italy to 11.6 in Spain. The relative risk (RR) of mortality, adjusted for age, sex and diabetes status was significantly higher in the UK (RR = 1.39, P = 0.02) compared with Italy (reference) and increased in association with age (RR = 1.60 for every 10 years older, P <0.001), diabetes as cause of end-stage renal disease (ESRD) (RR = 1.55, P < 0.001), male patients <65 years (RR = 1.29, P = 0.02) and peritoneal dialysis in the 12 months prior to starting haemodialysis (RR = 1.72, P = 0.06). Hospitalization for cardiovascular disease was highest in France and Germany (0.40 and 0.43 hospitalizations per patient year, respectively) and lowest in the UK (0.19), although cardiovascular comorbidity was similar in the UK and France. Hospitalization rates for vascular access-related infection ranged from 0.01 hospitalizations per patient year in Italy to 0.08 in the UK, consistent with the higher dialysis catheter use in the UK (25%) vs Italy (5%). Hospitalization risk was significantly higher in France than in other Euro-DOPPS countries and was significantly (P < 0.05) associated with prior peritoneal dialysis therapy, peripheral vascular disease, gastrointestinal bleeding in the prior 12 months, diabetes, cancer, cardiac disease, psychiatric disease and recent onset of ESRD (within 30 days of study entry). CONCLUSIONS: The large differences in haemodialysis practice and outcomes in the Euro-DOPPS countries suggest opportunities for improvement in patient care.

Characteristics of primary care physicians and their practices associated with mammography rates for older women.

BACKGROUND: Mammography screening rates are below national recommendations for older women. Understanding the relation between the characteristics of primary care physicians (PCPs) and mammography rates for older women can help to target screening improvement efforts. METHODS: Subjects were 2527 PCPs practicing in Michigan between 1997 and 1998. A cross-sectional design used Medicare data to identify women age 68 years or older in 1998 whom PCPs treated in 1997-1998 and to determine whether these women had a mammogram between 1996 and 1998. Eligible women were Medicare beneficiaries age 65 years or older by 1996, residing in Michigan from 1996 to 1998, without specified comorbidities likely to affect decisions regarding mammography. Correlations and multiple regressions examined the relation between this score and characteristics of both PCPs and their practice populations of older women. RESULTS: Mammography rates across physicians' practices ranged from 3-100% (mean = 59%, standard deviation = 17%). Five predictors accounted for 55% of the variance in mammography rates across practices. Higher mammography rates were found to be independently related to physicians who have: a lower mean age for female Medicare patients, a higher mean number of physicians billing for patients' care, a lower mean number of inpatient admissions, obstetrics/gynecology practices, and a higher mean education level in patient's zip code (beta weights >/= 0.25, P < 0.0001). CONCLUSIONS: PCPs vary substantially with regard to mammography rates for older women. Mammography rates vary more with the population of patients in physicians' practices than with commonly measured personal characteristics of physicians. Mammography rates should be adjusted for patient population to target individual PCPs with low mammography rates for interventions.

Personalized targeted mailing increases mammography among long-term noncompliant medicare beneficiaries: a randomized trial.

OBJECTIVES: The study purpose was to increase mammography screening among older women by identifying female Medicare beneficiaries without a recent mammogram and assesses the cost-effectiveness of a personalized targeted mailing encouraging them to have a mammogram. METHODS: A randomized paired controlled trial included 1229 pairs of women matched on zip code, race, and urban or rural county. Postintervention mammography claims were measured from November 1997 through December 1998. The subjects were female Medicare beneficiaries age > or = 70, living in Michigan for > or = 5 years, having no significant comorbidity likely to affect screening, and no mammogram for > or = 5 years. Intervention subjects received a personally addressed letter from the Medical Director of Michigan Medicare with materials emphasizing the individual's lack of use of the Medicare mammography screening benefit, reasons for screening, and how to be screened. RESULTS: Women who received the mailing were 60% more likely to have a subsequent mammogram (OR 1.6, P <0.005), with diagnostic mammograms increasing more than screening mammograms (2.8% vs. 0.8%). The absolute increase was greatest for women age 70 to 79, 10.6% in the intervention group versus 6.5% for controls, odds ratio 1.7 (P <0.02). A statewide Medicare intervention in Michigan would cost of 108,000 US dollars to 238,000 US dollars, producing 3500 to 4300 additional mammograms at 31 US dollars to 55 US dollars per additional mammogram. CONCLUSION: The intervention increased mammography among long-term noncompliant older women, particularly increasing diagnostic mammograms. This approach can be directly implemented in other states and nationally. It may also be useful for other preventive services.

5-Year mammography rates and associated factors for older women.

BACKGROUND: Major national interventions occurred in the early and mid-1990s to increase mammography screening rates among older women. The current study examined mammography utilization by older women during this period. Relation between mammography utilization and demographic measures and health care-related factors also were examined. METHODS: A cross-sectional design examined variations in mammography during the 5 years between 1993 to 1997 in a representative sample of 10,000 female Medicare beneficiaries in Michigan age >or= 65 years in 1993. Medicare and census data were used. Separate analyses were performed for having undergone any mammogram and, for the 5680 women who had undergone a mammogram, the number of mammograms. Relations were examined between mammography utilization and 15 demographic variables (e.g., age and African-American race) and health care-related variables (e.g., inpatient admissions and number of physicians involved in care). RESULTS: In the 5 years 43% of older women had no evidence of having undergone a mammogram. Those with any mammogram averaged 2.8 mammograms. Meaningful independent predictors of both having undergone a mammogram and having more than one mammogram were more physicians involved in care, fewer inpatient admissions, and younger age. Having undergone a mammogram also was found to be associated with seeing an obstetrician/gynecologist. CONCLUSIONS: Even with screening mammography as a covered benefit and after several national informational campaigns, the current study found that in 5 years, 60% of older women either had not undergone a mammogram or had undergone only 1. Intervention efforts should emphasize screening based on functional status, not age. This message should be targeted to physicians as well as to older women without claims for recent mammograms and who are likely to be in good health.

Cancers of the kidney and urinary tract in patients on dialysis for end-stage renal disease: analysis of data from the United States, Europe, and Australia and New Zealand.

Patients on maintenance dialysis have increased risk for cancer, especially in the kidney and urinary tract. In a retrospective cohort of 831,804 patients starting dialysis during 1980 to 1994 in the United States, Europe, or Australia and New Zealand, standardized incidence ratios (SIR) with 95% confidence intervals (CI) were calculated for kidney and bladder cancers. Risks for cancers of the kidney (SIR 3.6; CI 3.5 to 3.8) and bladder (SIR 1.5; CI 1.4 to 1.6) were increased, relatively more in younger than older patients and more in female patients (kidney: SIR 4.6, CI 4.3 to 4.9; bladder: SIR 2.7, CI 2.4 to 2.9) than male patients (kidney: SIR 3.2, CI 3.0 to 3.4; bladder: SIR 1.3, CI 1.2 to 1.3). SIR for kidney cancer were raised in all categories of primary renal disease, and for bladder cancer in all but diabetes and familial, hereditary diseases. Notably high SIR occurred in toxic nephropathies (chiefly analgesic nephropathy) and miscellaneous conditions (a category that includes Balkan nephropathy), the excess of kidney cancer in these conditions being urothelial in origin. SIR for kidney cancer rose significantly, and those for bladder cancer fell (not reaching significance) with time on dialysis. There was no association with type of dialysis. The pattern of increased risk for renal parenchymal cancer in dialysis patients is consistent with causation through acquired renal cystic disease and of urothelial cancers of the kidney and bladder with the carcinogenic effects of certain primary renal diseases.

Vascular access use in Europe and the United States: results from the DOPPS.

BACKGROUND: A direct broad-based comparison of vascular access use and survival in Europe (EUR) and the United States (US) has not been performed previously. Case series reports suggest that vascular access practices differ substantially in the US and EUR. We report on a representative study (DOPPS) which has used the same data collection protocol for> 6400 hemodialysis (HD) patients to compare vascular access use at 145 US dialysis units and 101 units in five EUR countries (France, Germany, Italy, Spain, and the United Kingdom). METHODS: Logistic analysis evaluated factors associated with native arteriovenous fistula (AVF) versus graft use or permanent access versus catheter use for prevalent and incident HD patients. Times to failure for AVF and graft were analyzed using Cox proportional hazards regression. RESULTS: AVF was used by 80% of EUR and 24% of US prevalent patients, and was significantly associated with younger age, male gender, lower body mass index, non-diabetic status, lack of peripheral vascular disease, and no angina. After adjusting for these factors, AVF versus graft use was still much higher in EUR than US (AOR=21, P < 0.0001). AVF use within facilities varied from 0 to 87% (median 21%) in the US, and 39 to 100% (median 83%) in EUR. For patients who were new to HD, access use was: 66% AVF in EUR versus 15% in US (AOR=39, P < 0.0001), 31% catheters in EUR vs. 60% in US, and 2% grafts in EUR vs. 24% in US. In addition, 25% of EUR and 46% of US incident patients did not have a permanent access placed prior to starting HD. In EUR, 84% of new HD patients had seen a nephrologist for> 30 days prior to ESRD compared with 74% in the US (P < 0.0001); pre-ESRD care was associated with increased odds of AVF versus graft use (AOR=1.9, P=0.01). New HD patients had a 1.8-fold greater odds (P=0.002) of starting HD with a permanent access if a facility's typical time from referral to access placement was < or =2 weeks. AVF use when compared to grafts was substantially lower (AOR=0.61, P=0.04) when surgery trainees assisted or performed access placements. When used as a patient's first access, AVF survival was superior to grafts regarding time to first failure (RR=0.53, P=0.0002), and AVF survival was longer in EUR compared with the US (RR=0.49, P=0.0005). AVF and grafts each displayed better survival if used when initiating HD compared with being used after patients began dialysis with a catheter. CONCLUSION: Large differences in vascular access use exist between EUR and the US, even after adjustment for patient characteristics. The results strongly suggest that a facility's preferences and approaches to vascular access practice are major determinants of vascular access use.