Arnold Barnett

The Relation between Experience and Outcome in Heart Transplantation

BACKGROUND Current policies related to organ transplantation in the United States are designed to ensure that centers and physicians with experience in transplantation perform these procedures. It is essential to confirm the validity of such policies, since they may limit access to transplantation services. METHODS To determine the relation between experience with heart transplantation and mortality after the procedure, we merged data from the registry of the International Society for Heart and Lung Transplantation with data from a survey that provided additional information about patients and transplantation centers. Our study included 1123 patients who received a heart transplant at one of 56 hospitals in the United States from 1984 through 1986. We used univariate and bivariate techniques, as well as logistic regression, to analyze our data. RESULTS We observed an institutional learning curve for heart transplantation. Patients who received one of a centers first five transplants had higher mortality rates than patients who received a subsequent transplant (20 percent vs. 12 percent; P = 0.002; relative risk = 2.2; 95 percent confidence interval, 1.6 to 3.4). In addition, we found a correlation between the training of key personnel on the transplantation team and mortality at new transplantation centers. For example, new centers staffed by cardiologists with previous training in heart transplantation had lower mortality rates among heart-transplant recipients than centers without experienced cardiologists (7 percent vs. 16 percent; P = 0.001; relative risk = 2.7; 95 percent confidence interval, 1.3 to 5.9). By contrast, the previous training of the surgeons who performed transplantations was not related to the mortality rate associated with the procedure. CONCLUSIONS Experience with heart transplantation is associated with a better outcome for patients after that procedure. Opportunities exist to refine transplantation policies on the basis of the experience of a center and its transplantation team and to develop similar policies for other forms of organ transplantation.

Landings at Logan Airport: Describing and Increasing Airport Capacity

To understand how greatly new computer-based Decision Support Systems can benefit air traffic control, we study air traffic delays for landing aircraft at Boston. First, we develop an empirical model for present day Landing Time Intervals (LTIs) between aircraft in terms of two factors that significantly affect them: the landing runway configuration and the weight-class categories of the aircraft. Next, we develop three increasingly rich models of Bostons terminal airspace and apply, on airflow data, sequencing algorithms meant to expedite the landing of incoming aircraft. Comparing sequences suggested by the algorithms to those now used by controllers, we estimate that better sequencing can reduce delays by 30% in some instances. However, such improvements must be balanced against the effects such algorithms would have on workloads and other aspects of the air traffic control environment.

The Decision to Perform Staging Laparotomy in Symptomatic Hodgkin's Disease

. This study defines patients with symptomatic Hodgkins disease for whom risks of staging laparotomy (LAP) outweigh benefits conferred by accurate knowledge of stage. From a database of more than 900 pathologically‐staged patients, probabilities of pathological stage are calculated for combinations of basic findings and lymphangiogram results. Decision‐making thresholds are defined at which results of treatment after LAP, taking operative mortality into account, are equivalent to immediate treatment appropriate to clinical stage. These thresholds are substantially altered by varying LAP mortality estimates, by assigning a false negative rate to LAP, and by considering uncertainty in treatment results. Fifty‐four combinations of findings are described for which immediate therapy with MOPP is justified; total nodal irradiation (TNI) is never indicated in B patients without prior LAP staging. Analysing 94 B patients who had LAP showed an appreciable number might have been spared this, particularly when uncertainty in treatment results is considered. With 0.03 (= 3%) uncertainty in treatment results, and 1% LAP mortality, LAP was not indicated in one in seven patients; nearly one third of patients should have immediate treatment with 3% LAP mortality. Threshold analysis can define those patients for whom risks of LAP outweigh benefits.

Selective incapacitation and the Philadelphia cohort data

The University of Pennsylvania study of delinquency in a Philadelphia birth cohort has been described byNewsweek as “perhaps the most influential piece of criminal justice research in the last decade” (March 23, 1981). Many have construed the findings as showing that, if imprisonment were focused on the minority of offenders with especially bad “prognoses,” the rate of crime could be reduced substantially. But others have taken the opposite view that the cohort data, far from endorsing such a “selective incapacitation” strategy, might actually provide strong evidence that such an approach is futile. Through some further analyses of the Philadelphia data, we attempt to clarify their policy implications.

America's Vietnam casualties: victims of a class war?

Analysis of data about the 58,000 Americans killed in Vietnam implies that affluent U.S. communities had only marginally lower casualty rates than the nation as a whole. Poor communities had only marginally higher rates. Data about the residential addresses of war casualties suggest that, within both large heterogeneous cities and wealthy suburbs, there was little relationship between neighborhood incomes and per capita Vietnam death rates. Such outcomes call into question a widespread belief that continues to influence U.S. policy discussions, namely, that American war deaths in Vietnam were overwhelmingly concentrated among the poor and working class.

On urban homicide: A statistical analysis

A statistical analysis is made of homicide rates in the 50 largest American cities for four different years. It is shown that differences in recent murder growth among the cities can largely be explained as typical random fluctuations about a common trend. It is also found that the changing age profile of the American people explains no more than ten percent of the increase in homicide since 1964. Several mathematical models for future homicide growth are proposed from the analysis, and under each the probability of death by murder and corresponding drop in life expectancy are estimated for individuals born now in each of the 50 cities.

A New Approach to Estimating the Probability of Winning the Presidency

As the 2000 election so vividly showed, it is Electoral College standings rather than national popular votes that determine who becomes President. But current pre-election polls focus almost exclusively on the popular vote. Here we present a method by which pollsters can achieve both point estimates and margins of error for a presidential candidates electoral-vote total. We use data from both the 2000 and 1988 elections to illustrate the approach. Moreover, we indicate that the sample sizes needed for reliable inferences are similar to those now used in popular-vote polling.

Prison Populations: A Projection Model

We propose a method for estimating prison populations under a variety of assumptions about sentencing policy and demographic structure. The approach builds on a familiar stochastic model of the individual criminal career. After a validation exercise involving Pennsylvania data, we use the model to make illustrative projections about Massachusetts, Utah and Florida. The results suggest that, while prison populations might stabilize everywhere during the early 1990s, only rarely will this situation signify an end to the growth in prison-capacity requirements.

Air Safety: End of the Golden Age?

First-World aviation has become so safe that a passenger who takes a domestic jet flight every day would on average go 36,000 years before succumbing to a fatal crash. But certain aerial dangers that were practically absent from the First World in the 1990’s might be poised for a resurgence. (Among these hazards are terrorism, mid-air collisions, and ground collisions.) We explore recent data about the mortality risk of air travel, and discuss the prospects for the years ahead.

Dose conversion from epoetin alfa to darbepoetin alfa for patients with chronic kidney disease receiving hemodialysis.

In his recent article in Pharmacotherapy, Dr. Shane Scott estimated an epoetin alfa:darbepoetin alfa dose conversion ratio (DCR) by comparing week 0 epoetin alfa doses with week 24 darbepoetin alfa doses from several studies in which patients receiving hemodialysis were converted from epoetin alfa to darbepoetin alfa. Based on evidence in Table 1 and Figure 3 of his article (replicated as Table 1 and Figure 1), Dr. Scott concluded that the DCR is nonlinear and increases with epoetin alfa dose and that there is “large interpatient variability” in DCRs. He therefore recommended that no fixed DCR between epoetin alfa and darbepoetin alfa be used when converting patients from epoetin alfa to darbepoetin alfa. This recommendation contrasts with the Centers for Medicare and Medicaid Services (CMS) recent adoption of a fixed DCR of 260:1. Dr. Scott’s findings are inconclusive for two reasons. First, he failed to provide statistically significant evidence against using a linear model with a 260:1 DCR. Second, his calculations about interpatient variation in DCRs posit that a patient’s appropriate epoetin alfa dose is constant over time, an assumption that is inconsistent with his data. His observations of a nonlinear DCR and large interpatient variability of DCRs could be construed as artifacts of variation over time in an individual patient’s dosing requirements. The current presumption of a fixed 260:1 DCR therefore should not be rejected on the basis of Dr. Scott’s evidence. Statistical Significance