Stuart G. Baker

Regression analysis for categorical variables with outcome subject to nonignorable nonresponse

Abstract We develop a log-linear model for categorical response subject to nonignorable nonresponse. The paper differs from Fay (1986) in its focus on estimation and hypothesis testing in a regression setting, as opposed to imputation in a multivariate setting. We present several new results concerning the existence of solutions on the boundary of the parameter space and the construction of confidence intervals for estimates. We illustrate the method by estimating the proportion of voters preferring Truman in a 1948 preelection poll (Mosteller, Hyman, McCarthy, Marks, and Truman 1949). Results may depend strongly on the model assumed for nonresponse; goodness-of-fit tests are available for comparing alternative models.

Wheezing during induction of general anesthesia in patients with and without asthma : a randomized, blinded trial

BackgroundPatients with asthma who require general anesthesia and tracheal intubation are at increased risk for the development of bronchospasm during induction. The incidence of wheezing during induction with different intravenously administered agents is unknown. A randomized, double-blinded prosp

A perfect correlate does not a surrogate make

BackgroundThere is common belief among some medical researchers that if a potential surrogate endpoint is highly correlated with a true endpoint, then a positive (or negative) difference in potential surrogate endpoints between randomization groups would imply a positive (or negative) difference in unobserved true endpoints between randomization groups. We investigate this belief when the potential surrogate and unobserved true endpoints are perfectly correlated within each randomization group.MethodsWe use a graphical approach. The vertical axis is the unobserved true endpoint and the horizontal axis is the potential surrogate endpoint. Perfect correlation within each randomization group implies that, for each randomization group, potential surrogate and true endpoints are related by a straight line. In this scenario the investigator does not know the slopes or intercepts. We consider a plausible example where the slope of the line is higher for the experimental group than for the control group.ResultsIn our example with unknown lines, a decrease in mean potential surrogate endpoints from control to experimental groups corresponds to an increase in mean true endpoint from control to experimental groups. Thus the potential surrogate endpoints give the wrong inference. Similar results hold for binary potential surrogate and true outcomes (although the notion of correlation does not apply). The potential surrogate endpointwould give the correct inference if either (i) the unknown lines for the two group coincided, which means that the distribution of true endpoint conditional on potential surrogate endpoint does not depend on treatment group, which is called the Prentice Criterion or (ii) if one could accurately predict the lines based on data from prior studies.ConclusionPerfect correlation between potential surrogate and unobserved true outcomes within randomized groups does not guarantee correct inference based on a potential surrogate endpoint. Even in early phase trials, investigators should not base conclusions on potential surrogate endpoints in which the only validation is high correlation with the true endpoint within a group.

Predictive accuracy of the Liverpool Lung Project risk model for stratifying patients for computed tomography screening for lung cancer: a case-control and cohort validation study.

BACKGROUND External validation of existing lung cancer risk prediction models is limited. Using such models in clinical practice to guide the referral of patients for computed tomography (CT) screening for lung cancer depends on external validation and evidence of predicted clinical benefit. OBJECTIVE To evaluate the discrimination of the Liverpool Lung Project (LLP) risk model and demonstrate its predicted benefit for stratifying patients for CT screening by using data from 3 independent studies from Europe and North America. DESIGN Case-control and prospective cohort study. SETTING Europe and North America. PATIENTS Participants in the European Early Lung Cancer (EUELC) and Harvard case-control studies and the LLP population-based prospective cohort (LLPC) study. MEASUREMENTS 5-year absolute risks for lung cancer predicted by the LLP model. RESULTS The LLP risk model had good discrimination in both the Harvard (area under the receiver-operating characteristic curve [AUC], 0.76 [95% CI, 0.75 to 0.78]) and the LLPC (AUC, 0.82 [CI, 0.80 to 0.85]) studies and modest discrimination in the EUELC (AUC, 0.67 [CI, 0.64 to 0.69]) study. The decision utility analysis, which incorporates the harms and benefit of using a risk model to make clinical decisions, indicates that the LLP risk model performed better than smoking duration or family history alone in stratifying high-risk patients for lung cancer CT screening. LIMITATIONS The model cannot assess whether including other risk factors, such as lung function or genetic markers, would improve accuracy. Lack of information on asbestos exposure in the LLPC limited the ability to validate the complete LLP risk model. CONCLUSION Validation of the LLP risk model in 3 independent external data sets demonstrated good discrimination and evidence of predicted benefits for stratifying patients for lung cancer CT screening. Further studies are needed to prospectively evaluate model performance and evaluate the optimal population risk thresholds for initiating lung cancer screening.

Cumulative Incidence of False-Positive Test Results in Lung Cancer Screening: A Randomized Trial

BACKGROUND Direct-to-consumer promotion of lung cancer screening has increased, especially low-dose computed tomography (CT). However, screening exposes healthy persons to potential harms, and cumulative false-positive rates for low-dose CT have never been formally reported. OBJECTIVE To quantify the cumulative risk that a person who participated in a 1- or 2-year lung cancer screening examination would receive at least 1 false-positive result, as well as rates of unnecessary diagnostic procedures. DESIGN Randomized, controlled trial of low-dose CT versus chest radiography. (ClinicalTrials.gov registration number: NCT00006382) SETTING Feasibility study for the ongoing National Lung Screening Trial. PATIENTS Current or former smokers, aged 55 to 74 years, with a smoking history of 30 pack-years or more and no history of lung cancer (n = 3190). INTERVENTION Random assignment to low-dose CT or chest radiography with baseline and 1 repeated annual screening; 1-year follow-up after the final screening. Randomization was centralized and stratified by age, sex, and study center. MEASUREMENTS False-positive screenings, defined as a positive screening with a completed negative work-up or 12 months or more of follow-up with no lung cancer diagnosis. RESULTS By using a Kaplan-Meier analysis, a persons cumulative probability of 1 or more false-positive low-dose CT examinations was 21% (95% CI, 19% to 23%) after 1 screening and 33% (CI, 31% to 35%) after 2. The rates for chest radiography were 9% (CI, 8% to 11%) and 15% (CI, 13% to 16%), respectively. A total of 7% of participants with a false-positive low-dose CT examination and 4% with a false-positive chest radiography had a resulting invasive procedure. LIMITATIONS Screening was limited to 2 rounds. Follow-up after the second screening was limited to 12 months. The false-negative rate is probably an underestimate. CONCLUSION Risks for false-positive results on lung cancer screening tests are substantial after only 2 annual examinations, particularly for low-dose CT. Further study of resulting economic, psychosocial, and physical burdens of these methods is warranted. PRIMARY FUNDING SOURCE National Cancer Institute.

Cumulative incidence of false-positive results in repeated, multimodal cancer screening

PURPOSE Multiple cancer screening tests have been advocated for the general population; however, clinicians and patients are not always well-informed of screening burdens. We sought to determine the cumulative risk of a false-positive screening result and the resulting risk of a diagnostic procedure for an individual participating in a multimodal cancer screening program. METHODS Data were analyzed from the intervention arm of the ongoing Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, a randomized controlled trial to determine the effects of prostate, lung, colorectal, and ovarian cancer screening on disease-specific mortality. The 68,436 participants, aged 55 to 74 years, were randomized to screening or usual care. Women received serial serum tests to detect cancer antigen 125 (CA-125), transvaginal sonograms, posteroanterior-view chest radiographs, and flexible sigmoidoscopies. Men received serial chest radiographs, flexible sigmoidoscopies, digital rectal examinations, and serum prostate-specific antigen tests. Fourteen screening examinations for each sex were possible during the 3-year screening period. RESULTS After 14 tests, the cumulative risk of having at least 1 false-positive screening test is 60.4% (95% CI, 59.8%–61.0%) for men, and 48.8% (95% CI, 48.1%–49.4%) for women. The cumulative risk after 14 tests of undergoing an invasive diagnostic procedure prompted by a false-positive test is 28.5% (CI, 27.8%–29.3%) for men and 22.1% (95% CI, 21.4%–22.7%) for women. CONCLUSIONS For an individual in a multimodal cancer screening trial, the risk of a false-positive finding is about 50% or greater by the 14th test. Physicians should educate patients about the likelihood of false positives and resulting diagnostic interventions when counseling about cancer screening.

Markers for early detection of cancer: Statistical guidelines for nested case-control studies

BackgroundRecently many long-term prospective studies have involved serial collection and storage of blood or tissue specimens. This has spurred nested case-control studies that involve testing some specimens for various markers that might predict cancer. Until now there has been little guidance in statistical design and analysis of these studies.MethodsTo develop statistical guidelines, we considered the purpose, the types of biases, and the opportunities for extracting additional information.ResultsThe following guidelines:(1) For the clearest interpretation, statistics should be based on false and true positive rates – not odds ratios or relative risks(2) To avoid overdiagnosis bias, cases should be diagnosed as a result of symptoms rather than on screening.(3) To minimize selection bias, the spectrum of control conditions should be the same in study and target screening populations.(4) To extract additional information, criteria for a positive test should be based on combinations of individual markers and changes in marker levels over time.(5) To avoid overfitting, the criteria for a positive marker combination developed in a training sample should be evaluated in a random test sample from the same study and, if possible, a validation sample from another study.(6) To identify biomarkers with true and false positive rates similar to mammography, the training, test, and validation samples should each include at least 110 randomly selected subjects without cancer and 70 subjects with cancer.ConclusionThese guidelines ensure good practice in the design and analysis of nested case-control studies of early detection biomarkers.

MARGINAL REGRESSION FOR REPEATED BINARY DATA WITH OUTCOME SUBJECT TO NON-IGNORABLE NON-RESPONSE

Using a model that accounts for non-ignorable non-response, we analyzed data from the Muscatine Risk Factor Study (Woolson and Clarke, 1984, Journal of the Royal Statistical Society, Series A 147, 87-99) on the effects of gender and age on obesity in schoolchildren. The methodology is related to that of Diggle and Kenward (1994, Applied Statistics 43, 49-93), except that the repeated data are binary, not continuous, and the non-response occurs in various patterns, not just dropouts. We found strong evidence that non-response was non-ignorable. In addition, we found that the proportion of children who were obese differed significantly with gender and increased with age.

Correction: A simple method for analyzing data from a randomized trial with a missing binary outcome

In this article [1], equation (8) is incorrect because it omitted the covariance terms. Let h denote the number of strata, so s = 1,2,...,h. Let T denote transpose, • denote matrix product, and Diagonal Matrix [vector] denote a matrix of all 0s except for vector on the diagonal. The correct formula is where In our example, the effect of the correction was negligible; the corrected estimated standard error was the same to two significant digits as the incorrect value. Also for clarification, we note that in the sentence after (11), it is an assumption that, within stratum s, the difference, Δs, does not depend on the unobserved covariate x.

Analysis of Survival Data from a Randomized Trial with All-or-None Compliance: Estimating the Cost-Effectiveness of a Cancer Screening Program

Abstract Many randomized cancer screening trials involve all-or-none compliance. Some subjects randomized to an offer of screening refuse screening, and some subjects randomized to no offer of screening obtain screening outside the trial. The primary analysis to test whether or not cancer screening reduces cancer mortality is by intent-to-treat. To estimate the cost-effectiveness of screening, it is necessary to adjust for all-or-none compliance. Heretofore, adjustments for all-or-none compliance have been limited to a fixed-time endpoint. Estimating cost-effectiveness as dollars per life year saved requires an extension to the analysis of yearly survival data. In general, this involves modeling both the hazard for death from cancer and death from competing risk. Unconstrained estimates and variances can be written in closed-form notation. For the four yearly breast cancer screens with physical examination and mammography in the Health Insurance Plan of Greater New York study, the estimated cost-effective...