The performance of multiple imputation for missing covariate data within the context of regression relative survival analysis

Multi-metric comparison of machine learning imputation methods with application to breast cancer survival

Handling missing data in clinical prognostic studies is an essential yet challenging task. This study aimed to provide a comprehensive assessment of the effectiveness and reliability of different machine learning (ML) imputation methods across various analytical perspectives. Specifically, it focused on three distinct classes of performance metrics used to evaluate ML imputation methods: post-imputation bias of regression estimates, post-imputation predictive accuracy, and substantive model-free metrics. As an illustration, we applied data from a real-world breast cancer survival study. This comprehensive approach aimed to provide a thorough assessment of the effectiveness and reliability of ML imputation methods across various analytical perspectives. A simulated dataset with 30% Missing At Random (MAR) values was used. A number of single imputation (SI) methods - specifically KNN, missMDA, CART, missForest, missRanger, missCforest - and multiple imputation (MI) methods - specifically miceCART and miceRF - were evaluated. The performance metrics used were Gower's distance, estimation bias, empirical standard error, coverage rate, length of confidence interval, predictive accuracy, proportion of falsely classified (PFC), normalized root mean squared error (NRMSE), AUC, and C-index scores. The analysis revealed that in terms of Gower's distance, CART and missForest were the most accurate, while missMDA and CART excelled for binary covariates; missForest and miceCART were superior for continuous covariates. When assessing bias and accuracy in regression estimates, miceCART and miceRF exhibited the least bias. Overall, the various imputation methods demonstrated greater efficiency than complete-case analysis (CCA), with MICE methods providing optimal confidence interval coverage. In terms of predictive accuracy for Cox models, missMDA and missForest had superior AUC and C-index scores. Despite offering better predictive accuracy, the study found that SI methods introduced more bias into the regression coefficients compared to MI methods. This study underlines the importance of selecting appropriate imputation methods based on study goals and data types in time-to-event research. The varying effectiveness of methods across the different performance metrics studied highlights the value of using advanced machine learning algorithms within a multiple imputation framework to enhance research integrity and the robustness of findings.

Five-Feature Models to Predict Preeclampsia Onset Time From Electronic Health Record Data: Development and Validation Study

BACKGROUND

 Preeclampsia is a potentially fatal complication during pregnancy, characterized by high blood pressure and the presence of excessive proteins in the urine. Due to its complexity, the prediction of preeclampsia onset is often difficult and inaccurate.

OBJECTIVE

 This study aimed to create quantitative models to predict the onset gestational age of preeclampsia using electronic health records.

METHODS

 We retrospectively collected 1178 preeclamptic pregnancy records from the University of Michigan Health System as the discovery cohort, and 881 records from the University of Florida Health System as the validation cohort. We constructed 2 Cox-proportional hazards models: 1 baseline model using maternal and pregnancy characteristics, and the other full model with additional laboratory findings, vitals, and medications. We built the models using 80% of the discovery data, tested the remaining 20% of the discovery data, and validated with the University of Florida data. We further stratified the patients into high- and low-risk groups for preeclampsia onset risk assessment.

RESULTS

 The baseline model reached Concordance indices of 0.64 and 0.61 in the 20% testing data and the validation data, respectively, while the full model increased these Concordance indices to 0.69 and 0.61, respectively. For preeclampsia diagnosed at 34 weeks, the baseline and full models had area under the curve (AUC) values of 0.65 and 0.70, and AUC values of 0.69 and 0.70 for preeclampsia diagnosed at 37 weeks, respectively. Both models contain 5 selective features, among which the number of fetuses in the pregnancy, hypertension, and parity are shared between the 2 models with similar hazard ratios and significant P values. In the full model, maximum diastolic blood pressure in early pregnancy was the predominant feature.

CONCLUSIONS

 Electronic health records data provide useful information to predict the gestational age of preeclampsia onset. Stratification of the cohorts using 5-predictor Cox-proportional hazards models provides clinicians with convenient tools to assess the onset time of preeclampsia in patients.

Identification of risk factors on rabies vaccine efficacy from censored data: Pre-travel tests for dogs and cats from Yaoundé (2005-2015)

Little research is available on acquired immunity to rabies in dogs and cats from Central Africa, particularly regarding the legal movements of pets. Movement of domestic animals from rabies-endemic countries like Cameroon to rabies free areas poses one of the main risks for rabies introduction into rabies-free areas. Thus, the aim of this study was to assess the effect of various risk factors on rabies vaccine efficacy in Cameroonian. Since the dependent variable, rabies neutralizing titres, were censored from above (right-censoring), Generalized Additive Model for Location, Scale and Shape (GAMLSS) was used in the analysis. Overall, 85.7% of dogs and 100% of cats had titres greater than or equal to 0.5 IU/mL, which is considered protective. Additionally, compared to cats, the value of the rabies-neutralizing serum titres in dogs was on average smaller by 2.3 IU/mL. For each additional year of age, the value of the rabies-neutralizing serum titre, on average, increased by approximately 0.14 IU/mL. Finally, for each 30 additional days between the date of the last rabies vaccination and the date of the sampling, the value the rabies neutralizing titre, on average, decreased by approximately 0.10 IU/mL, given the species and age at sampling were equivalent. These results are useful for assessing risk and improving surveillance to prevent the introduction of rabies into a country via the international movement of animals.

The Effect of Missing Item Data on the Relative Predictive Accuracy of Correctional Risk Assessment Tools

Missing data are pervasive in risk assessment but their impact on predictive accuracy has largely been unexplored. Common techniques for handling missing risk data include summing available items or proration; however, multiple imputation is a more defensible approach that has not been methodically tested against these simpler techniques. We compared the validity of these three missing data techniques across six conditions using STABLE-2007 (N = 4,286) and SARA-V2 (N = 455) assessments from men on community supervision in Canada. Condition 1 was the observed data (low missingness), and Conditions 2 to 6 were generated missing data conditions, whereby 1% to 50% of items per case were randomly deleted in 10% increments. Relative predictive accuracy was unaffected by missing data, and simpler techniques performed just as well as multiple imputation, but summed totals underestimated absolute risk. The current study therefore provides empirical justification for using proration when data are missing within a sample.

Dealing with missing information on covariates for excess mortality hazard regression models - Making the imputation model compatible with the substantive model

Missing data is a common issue in epidemiological databases. Among the different ways of dealing with missing data, multiple imputation has become more available in common statistical software packages. However, the incompatibility between the imputation and substantive model, which can arise when the associations between variables in the substantive model are not taken into account in the imputation models or when the substantive model is itself nonlinear, can lead to invalid inference. Aiming at analysing population-based cancer survival data, we extended the multiple imputation substantive model compatible-fully conditional specification (SMC-FCS) approach, proposed by Bartlett et al. in 2015 to accommodate excess hazard regression models. The proposed approach was compared with the standard fully conditional specification multiple imputation procedure and with the complete-case analysis using a simulation study. The SMC-FCS approach produced unbiased estimates in both scenarios tested, while the fully conditional specification produced biased estimates and poor empirical coverages probabilities. The SMC-FCS algorithm was then used for handling missing data in the evaluation of socioeconomic inequalities in survival from colorectal cancer patients diagnosed in the North Region of Portugal. The analysis using SMC-FCS showed a clearer trend in higher excess hazards for patients coming from more deprived areas. The proposed algorithm was implemented in R software and is presented as Supplementary Material.

Evaluation of Multiple Imputation with Large Proportions of Missing Data: How Much Is Too Much?

Background:

Multiple Imputation (MI) is known as an effective method for handling missing data in public health research. However, it is not clear that the method will be effective when the data contain a high percentage of missing observations on a variable.

Methods:

Using data from “Predictive Study of Coronary Heart Disease” study, this study examined the effectiveness of multiple imputation in data with 20% missing to 80% missing observations using absolute bias (|bias|) and Root Mean Square Error (RMSE) of MI measured under Missing Completely at Random (MCAR), Missing at Random (MAR), and Not Missing at Random (NMAR) assumptions.

Results:

The |bias| and RMSE of MI was much smaller than of the results of CCA under all missing mechanisms, especially with a high percentage of missing. In addition, the |bias| and RMSE of MI were consistent regardless of increasing imputation numbers from M=10 to M=50. Moreover, when comparing imputation mechanisms, MCMC method had universally smaller |bias| and RMSE than those of Regression method and Predictive Mean Matching method under all missing mechanisms.

Conclusion:

As missing percentages become higher, using MI is recommended, because MI produced less biased estimates under all missing mechanisms. However, when large proportions of data are missing, other things need to be considered such as the number of imputations, imputation mechanisms, and missing data mechanisms for proper imputation.

When Is a Complete-Case Approach to Missing Data Valid? The Importance of Effect-Measure Modification

When estimating causal effects, careful handling of missing data is needed to avoid bias. Complete-case analysis is commonly used in epidemiologic analyses. Previous work has shown that covariate-stratified effect estimates from complete-case analysis are unbiased when missingness is independent of the outcome conditional on the exposure and covariates. Here, we assess the bias of complete-case analysis for adjusted marginal effects when confounding is present under various causal structures of missing data. We show that estimation of the marginal risk difference requires an unbiased estimate of the unconditional joint distribution of confounders and any other covariates required for conditional independence of missingness and outcome. The dependence of missing data on these covariates must be considered to obtain a valid estimate of the covariate distribution. If none of these covariates are effect-measure modifiers on the absolute scale, however, the marginal risk difference will equal the stratified risk differences and the complete-case analysis will be unbiased when the stratified effect estimates are unbiased. Estimation of unbiased marginal effects in complete-case analysis therefore requires close consideration of causal structure and effect-measure modification.

Individual participant data meta-analysis of intervention studies with time-to-event outcomes: A review of the methodology and an applied example

Many randomized trials evaluate an intervention effect on time-to-event outcomes. Individual participant data (IPD) from such trials can be obtained and combined in a so-called IPD meta-analysis (IPD-MA), to summarize the overall intervention effect. We performed a narrative literature review to provide an overview of methods for conducting an IPD-MA of randomized intervention studies with a time-to-event outcome. We focused on identifying good methodological practice for modeling frailty of trial participants across trials, modeling heterogeneity of intervention effects, choosing appropriate association measures, dealing with (trial differences in) censoring and follow-up times, and addressing time-varying intervention effects and effect modification (interactions).We discuss how to achieve this using parametric and semi-parametric methods, and describe how to implement these in a one-stage or two-stage IPD-MA framework. We recommend exploring heterogeneity of the effect(s) through interaction and non-linear effects. Random effects should be applied to account for residual heterogeneity of the intervention effect. We provide further recommendations, many of which specific to IPD-MA of time-to-event data from randomized trials examining an intervention effect.We illustrate several key methods in a real IPD-MA, where IPD of 1225 participants from 5 randomized clinical trials were combined to compare the effects of Carbamazepine and Valproate on the incidence of epileptic seizures.

Variation in treatment and survival of older patients with non-metastatic breast cancer in five European countries: a population-based cohort study from the EURECCA Breast Cancer Group

BACKGROUND

Older patients are poorly represented in breast cancer research and guidelines do not provide evidence based recommendations for this specific group. We compared treatment strategies and survival outcomes between European countries and assessed whether variance in treatment patterns may be associated with variation in survival.

METHODS

Population-based study including patients aged ≥ 70 with non-metastatic BC from cancer registries from the Netherlands, Belgium, Ireland, England and Greater Poland. Proportions of local and systemic treatments, five-year relative survival and relative excess risks (RER) between countries were calculated.

RESULTS

In total, 236,015 patients were included. The proportion of stage I BC receiving endocrine therapy ranged from 19.6% (Netherlands) to 84.6% (Belgium). The proportion of stage III BC receiving no breast surgery varied between 22.0% (Belgium) and 50.8% (Ireland). For stage I BC, relative survival was lower in England compared with Belgium (RER 2.96, 95%CI 1.30-6.72, P < .001). For stage III BC, England, Ireland and Greater Poland showed significantly worse relative survival compared with Belgium.

CONCLUSIONS

There is substantial variation in treatment strategies and survival outcomes in elderly with BC in Europe. For early-stage BC, we observed large variation in endocrine therapy but no variation in relative survival, suggesting potential overtreatment. For advanced BC, we observed higher survival in countries with lower proportions of omission of surgery, suggesting potential undertreatment.

Seed dispersal syndromes in the Madagascan flora: the unusual importance of primates

Madagascar is one of the most threatened biodiversity hotspots, and protection of its biodiversity is becoming increasingly urgent as deforestation of the island continues. For the long-term success of conservation efforts it is essential that key ecological processes, such as seed dispersal, are protected and restored. Therefore, the identification of ecological gaps is a vital task. For Madagascar, only little is known about plant–animal interactions, and traditional methods of ecological research are too time-consuming to provide crucial information about breakdowns in these interactions. To identify likely dispersal gaps we therefore used a theoretical approach to analyse plant–disperser interactions in Madagascar. We used data science tools to impute missing data on relevant plant traits to subsequently predict the most likely dispersal agents for each of Madagascar's endemic plant species. We found that 38% of the endemic species (N = 8,784) are endozoochorous, and among these 26–41% display a primate syndrome and 17–19% a bird syndrome (depending on the definition of syndromes). This lower percentage of endozoochorous species and higher percentage of species with a primate syndrome in Madagascar compared to other tropical areas reflects the unusual disperser guild on the island. Only five bird species but 20 lemur species are frugivorous, and 16 of those lemur species are currently threatened with extinction. The disappearance of frugivorous lemurs would significantly change the vegetation dynamics of Madagascar's ecosystems, and a high proportion of Madagascar's endemic plants would enter an extinction vortex.

Utility of Adjuvant Chemotherapy After Neoadjuvant Chemoradiation and Esophagectomy for Esophageal Cancer

OBJECTIVE

To determine whether adjuvant chemotherapy (AC) after neoadjuvant chemoradiation and esophagectomy is associated with improved overall survival for patients with locally advanced esophageal cancer, and to evaluate how pathologic disease response to neoadjuvant treatment impacts this effect.

BACKGROUND

Neoadjuvant chemoradiation is currently the preferred management approach for locoregional esophageal cancer. Although there is interest in the use of AC, the benefit of systemic therapy after neoadjuvant chemoradiation and esophagectomy is unclear.

METHODS

Retrospective cohort study of patients with esophageal cancer treated with neoadjuvant chemoradiation and esophagectomy in the National Cancer Data Base (2006-2012).

RESULTS

Among 3592 patients with esophageal cancer (84.7% adenocarcinoma, 15.2% squamous cell carcinoma), 335 (9.3%) were treated with AC. AC was not associated with a significantly lower risk of death among patients with no residual disease (ypT0N0) or residual non-nodal disease (ypT+N0). Among patients with residual nodal disease (ypTanyN+), AC was associated with a 30% lower risk of death in the overall cohort [hazard ratio (HR) 0.70, (0.57-0.85)] and among those with adenocarcinoma [HR 0.69 (0.57-0.85)]. Using a 90-day postoperative landmark, findings were similar. Among patients with postoperative length of stay ≤10 days and no unplanned readmission, AC was associated with approximately 40% lower risk of death among patients with residual nodal disease [overall cohort, HR 0.63 (0.48-0.84); adenocarcinoma, HR 0.66 (0.49-0.88)].

CONCLUSIONS

AC after neoadjuvant chemoradiation and esophagectomy is associated with improved survival in patients with residual nodal disease. Our findings suggest AC may provide additional benefit for esophageal cancer patients, and merits further investigation.

Factors related to the relative survival of patients with diffuse large B-cell lymphoma in a population-based study in France: does socio-economic status have a role?

The survival of patients with diffuse large B-cell lymphoma has increased during the last decade as a result of addition of anti-CD20 to anthracycline-based chemotherapy. Although the trend is encouraging, there are persistent differences in survival within and between the USA and European countries suggesting that non-biological factors play a role. Our aim was to investigate the influence of such factors on relative survival of patients with diffuse large B-cell lymphoma. We conducted a retrospective, multicenter, registry-based study in France on 1165 incident cases of diffuse large B-cell lymphoma between 2002 and 2008. Relative survival analyses were performed and missing data were controlled with the multiple imputation method. In a multivariate analysis, adjusted for age, sex and International Prognostic Index, we confirmed that time period was associated with a better 5-year relative survival. The registry area, the medical specialty of the care department (onco-hematology versus other), the time to travel to the nearest teaching hospital, the place of treatment (teaching versus not-teaching hospital -borderline significance), a comorbidity burden and marital status were independently associated with the 5-year relative survival. Adjusted for first-course treatment, inclusion in a clinical trial and treatment discussion in a multidisciplinary meeting were strongly associated with a better survival outcome. In contrast, socio-economic status (determined using the European Deprivation Index) was not associated with outcome. Despite therapeutic advances, various non-biological factors affected the relative survival of patients with diffuse large B-cell lymphoma. The notion of lymphoma-specific expertise seems to be essential to achieve optimal care management and reopens the debate regarding centralization of these patients’ care in hematology/oncology departments.

Adequacy of the National Quality Forum's Colon Cancer Adjuvant Chemotherapy Quality Metric: Is 4 Months Soon Enough?

OBJECTIVE

To ascertain whether the National Quality Forum-endorsed time interval for adjuvant chemotherapy (AC) initiation optimizes patient outcome.

BACKGROUND

Delayed AC initiation for stage III colon cancer is associated with worse survival and the focus of a National Quality Forum quality metric (<4 months among patients aged <80 years).

METHODS

Observational cohort study of patients with stage III colon cancer younger than 80 years within the National Cancer Data Base (2003-2010). The primary outcome was 5-year overall survival evaluated using multivariate Cox proportional hazards regression. Aggregate survival estimates for historical surgery-only controls from pooled National Surgical Adjuvant Breast and Bowel Project trial data were also used.

RESULTS

Among 51,331 patients (60.8 ± 11.6 years, 50.2% males, and 77.3% white), 76.3% received standard AC (≤ 2 months) and 21.6% delayed (>2 and <4 months) AC. Earlier AC was associated with better 5-year overall survival [standard AC, 69.8%; delayed AC, 62.0%; late AC (4-6 months), 51.4%; log-rank, P < 0.001]. The survival rate after late AC was similar to surgery alone (51.1%; Wilcoxon rank sum, P = 0.10). Compared with late AC, standard AC (hazard ratio, 0.62; 95% confidence interval, 0.54-0.72) and delayed AC (hazard ratio, 0.77; 95% confidence interval, 0.66-0.89) significantly decreased risk of death. Risk of death was also lower for standard AC compared to delayed AC (hazard ratio, 0.81; 95% confidence interval, 0.77-0.86).

CONCLUSIONS

One in 5 patients with stage III colon cancer initiates AC within the National Quality Forum-endorsed interval, but does not derive the full benefit. These data support strengthening current quality improvement initiatives and colon cancer treatment guidelines to encourage AC initiation within 2 months of resection when possible, but not beyond 4 months.

Subsample ignorable likelihood for accelerated failure time models with missing predictors

Missing values in predictors are a common problem in survival analysis. In this paper, we review estimation methods for accelerated failure time models with missing predictors, and apply a new method called subsample ignorable likelihood (IL) Little and Zhang (J R Stat Soc 60:591-605, 2011) to this class of models. The approach applies a likelihood-based method to a subsample of observations that are complete on a subset of the covariates, chosen based on assumptions about the missing data mechanism. We give conditions on the missing data mechanism under which the subsample IL method is consistent, while both complete-case analysis and ignorable maximum likelihood are inconsistent. We illustrate the properties of the proposed method by simulation and apply the method to a real dataset.

Risk-adjusted pathologic margin positivity rate as a quality indicator in rectal cancer surgery

PURPOSE

Margin positivity after rectal cancer resection is associated with poorer outcomes. We previously developed an instrument for calculating hospital risk-adjusted margin positivity rate (RAMP) that allows identification of performance-based outliers and may represent a rectal cancer surgery quality metric.

METHODS

This was an observational cohort study of patients with rectal cancer within the National Cancer Data Base (2003 to 2005). Hospital performance was categorized as low outlier (better than expected), high outlier (worse than expected), or non-RAMP outlier using standard observed-to-expected methodology. The association between outlier status and overall risk of death at 5 years was evaluated using Cox shared frailty modeling.

RESULTS

Among 32,354 patients with cancer (mean age, 63.8 +/-13.2 years; 56.7% male; 87.3% white) treated at 1,349 hospitals (4.9% high outlier, 0.7% low outlier), 5.6% of patients were treated at high outliers and 3.0% were treated at low outliers. Various structural (academic status and volume), process (pathologic nodal evaluation and neoadjuvant radiation therapy use), and outcome (sphincter preservation, readmission, and 30-day postoperative mortality) measures were significantly associated with outlier status. Five-year overall survival was better at low outliers (79.9%) compared with high outliers (64.9%) and nonoutliers (68.9%; log-rank test, P < .001). Risk of death was lower at low outliers compared with high outliers (hazard ratio [HR], 0.61; 95% CI, 0.50 to 0.75) and nonoutliers (HR, 0.69; 95% CI, 0.57 to 0.83). Risk of death was higher at high outliers compared with nonoutliers (HR, 1.12; 95% CI, 1.03 to 1.23).

CONCLUSION

Hospital RAMP outlier status is a rectal cancer surgery composite metric that reliably captures hospital quality across all levels of care and could be integrated into existing quality improvement initiatives for hospital performance.

Nonlinear multiple imputation for continuous covariate within semiparametric Cox model: application to HIV data in Senegal

Multiple imputation is commonly used to impute missing covariate in Cox semiparametric regression setting. It is to fill each missing data with more plausible values, via a Gibbs sampling procedure, specifying an imputation model for each missing variable. This imputation method is implemented in several softwares that offer imputation models steered by the shape of the variable to be imputed, but all these imputation models make an assumption of linearity on covariates effect. However, this assumption is not often verified in practice as the covariates can have a nonlinear effect. Such a linear assumption can lead to a misleading conclusion because imputation model should be constructed to reflect the true distributional relationship between the missing values and the observed values. To estimate nonlinear effects of continuous time invariant covariates in imputation model, we propose a method based on B-splines function. To assess the performance of this method, we conducted a simulation study, where we compared the multiple imputation method using Bayesian splines imputation model with multiple imputation using Bayesian linear imputation model in survival analysis setting. We evaluated the proposed method on the motivated data set collected in HIV-infected patients enrolled in an observational cohort study in Senegal, which contains several incomplete variables. We found that our method performs well to estimate hazard ratio compared with the linear imputation methods, when data are missing completely at random, or missing at random.

Progression of coronary calcium and incident coronary heart disease events: MESA (Multi-Ethnic Study of Atherosclerosis)

OBJECTIVES

The study examined whether progression of coronary artery calcium (CAC) is a predictor of future coronary heart disease (CHD) events.

BACKGROUND

CAC predicts CHD events and serial measurement of CAC has been proposed to evaluate atherosclerosis progression.

METHODS

We studied 6,778 persons (52.8% female) aged 45 to 84 years from the MESA (Multi-Ethnic Study of Atherosclerosis) study. A total of 5,682 persons had baseline and follow-up CAC scans approximately 2.5 ± 0.8 years apart; multiple imputation was used to account for the remainder (n = 1,096) missing follow-up scans. Median follow-up duration from the baseline was 7.6 (max = 9.0) years. CAC change was assessed by absolute change between baseline and follow-up CAC. Cox proportional hazards regression providing hazard ratios (HRs) examined the relation of change in CAC with CHD events, adjusting for age, gender, ethnicity, baseline calcium score, and other risk factors.

RESULTS

A total of 343 and 206 hard CHD events occurred. The annual change in CAC averaged 24.9 ± 65.3 Agatston units. Among persons without CAC at baseline (n = 3,396), a 5-unit annual change in CAC was associated with an adjusted HR (95% Confidence Interval) of 1.4 (1.0 to 1.9) for total and 1.5 (1.1 to 2.1) for hard CHD. Among those with CAC >0 at baseline, HRs (per 100 unit annual change) were 1.2 (1.1 to 1.4) and 1.3 (1.1 to 1.5), respectively. Among participants with baseline CAC, those with annual progression of ≥300 units had adjusted HRs of 3.8 (1.5 to 9.6) for total and 6.3 (1.9 to 21.5) for hard CHD compared to those without progression.

CONCLUSIONS

Progression of CAC is associated with an increased risk for future hard and total CHD events.

Dealing with missing data in the Center for Epidemiologic Studies Depression self-report scale: a study based on the French E3N cohort

BACKGROUND

The Center for Epidemiologic Studies - Depression scale (CES-D) is a validated tool commonly used to screen depressive symptoms. As with any self-administered questionnaire, missing data are frequently observed and can strongly bias any inference. The objective of this study was to investigate the best approach for handling missing data in the CES-D scale.

METHODS

Among the 71,412 women from the French E3N prospective cohort (Etude Epidémiologique auprès des femmes de la Mutuelle Générale de l'Education Nationale) who returned the questionnaire comprising the CES-D scale in 2005, 45% had missing values in the scale. The reasons for failure to complete certain items were investigated by semi-directive interviews on a random sample of 204 participants. The prevalence of high depressive symptoms (score ≥ 16, hDS) was estimated after applying various methods for ignorable missing data including multiple imputation using imputation models with CES-D items with or without covariates. The accuracy of imputation models was investigated. Various scenarios of nonignorable missing data mechanisms were investigated by a sensitivity analysis based on the mixture modelling approach.

RESULTS

The interviews showed that participants were not reluctant to answer the CES-D scale. Possible reasons for nonresponse were identified. The prevalence of hDS among complete responders was 26.1%. After multiple imputation, the prevalence was 28.6%, 29.8% and 31.7% for women presenting up to 4, 10 and 20 missing values, respectively. The estimates were robust to the various imputation models investigated and to the scenarios of nonignorable missing data.

CONCLUSIONS

The CES-D scale can easily be used in large cohorts even in the presence of missing data. Based on the results from both a qualitative study and a sensitivity analysis under various scenarios of missing data mechanism in a population of women, missing data mechanism does not appear to be nonignorable and estimates are robust to departures from ignorability. Multiple imputation is recommended to reliably handle missing data in the CES-D scale.

Excess mortality and morbidity in patients surviving infective endocarditis

BACKGROUND

Mortality and morbidity associated with infective endocarditis may extend beyond successful treatment. The primary objective was to analyze rates, temporal changes, and predictors of excess mortality in patients surviving the acute phase of endocarditis. The secondary objective was to determine the rate of recurrence and the need for late cardiac surgery.

METHODS

An observational cohort study was conducted at a university-affiliated tertiary medical center, among 328 patients who survived the active phase of endocarditis. We used age-, sex-, and calendar year-specific mortality hazard rates of the Bouches-du-Rhone French district population to calculate expected survival and excess mortality. The risk of recurrence and late valve surgery was also assessed.

RESULT

Compared with expected survival, patients surviving a first episode of endocarditis had significantly worse outcomes (P = .001). The relative survival rates at 1, 3, and 5 years were 92% (95% CI, 88%-95%), 86% (95% CI, 77%-92%), and 82% (95% CI, 59%-91%), respectively. This excess mortality was observed during the entire follow-up period but was the highest during the first year after hospital discharge. Most of the recurrences and late cardiac surgeries also occurred during this period. Women exhibited a higher risk of age-adjusted excess mortality (adjusted excess hazard ratio, 2.0; 95% CI, 1.05-3.82; P = .03). Comorbidity index, recurrence of endocarditis, and history of an aortic valve endocarditis in women were independent predictors of excess mortality.

CONCLUSIONS

These results justify close monitoring of patients after successful treatment of endocarditis, at least during the first year. Special attention should be paid to women with aortic valve damage.

Comparative assessment of the safety and effectiveness of radiofrequency ablation among elderly medicare beneficiaries with hepatocellular carcinoma

BACKGROUND

Radiofrequency ablation (RFA) use among patients with hepatocellular carcinoma (HCC) has increased dramatically over the last decade, but assessments outside specialized centers are lacking. This population-based study was intended to evaluate the safety and effectiveness of RFA when used to treat HCC.

METHODS

A cohort study of HCC patients (diagnosed 2002-2005) was performed using linked Surveillance, Epidemiology, and End Results (SEER)-Medicare data. Early (≤90-day) mortality and readmission as well as survival among patients undergoing RFA, resection, or no treatment were compared using multivariate and propensity score adjusted Poisson and Cox regression models.

RESULTS

Of 2631 patients (mean age 76.1±6.1 years, 65.9% male), 16% underwent RFA (49.6%) or resection (50.4%). Early mortality (13.6 vs. 18.7%, P=.16) and readmission (34.5 vs. 32.1%, P=.60) rates were similar among RFA and resection patients. The 1-year survival after RFA and resection was similar (72.2 vs. 79.7%, P=.18), but beyond 3 years there was a survival benefit among patients undergoing resection (39.2 vs. 58.0%, P<.001). Patients treated with RFA as a sole therapeutic intervention in the 1st year had a similar hazard of death compared with untreated patients (hazard ratio [HR] 0.84, 95% confidence interval [95% CI] 0.54-1.33).

CONCLUSIONS

In the general community, patients treated with RFA have a similar risk of early adverse events compared with those treated with resection with no clear survival benefit when used as a sole intervention. Although RFA has been described as a safe and effective treatment for HCC at specialized centers, this experience may not extrapolate to the general community and requires further evaluation.

Using causal diagrams to guide analysis in missing data problems

Estimating causal effects from incomplete data requires additional and inherently untestable assumptions regarding the mechanism giving rise to the missing data. We show that using causal diagrams to represent these additional assumptions both complements and clarifies some of the central issues in missing data theory, such as Rubin's classification of missingness mechanisms (as missing completely at random (MCAR), missing at random (MAR) or missing not at random (MNAR)) and the circumstances in which causal effects can be estimated without bias by analysing only the subjects with complete data. In doing so, we formally extend the back-door criterion of Pearl and others for use in incomplete data examples. These ideas are illustrated with an example drawn from an occupational cohort study of the effect of cosmic radiation on skin cancer incidence.

STS database risk models: predictors of mortality and major morbidity for lung cancer resection

BACKGROUND

The aim of this study is to create models for perioperative risk of lung cancer resection using the STS GTDB (Society of Thoracic Surgeons General Thoracic Database).

METHODS

The STS GTDB was queried for all patients treated with resection for primary lung cancer between January 1, 2002 and June 30, 2008. Three separate multivariable risk models were constructed (mortality, major morbidity, and composite mortality or major morbidity).

RESULTS

There were 18,800 lung cancer resections performed at 111 participating centers. Perioperative mortality was 413 of 18,800 (2.2%). Composite major morbidity or mortality occurred in 1,612 patients (8.6%). Predictors of mortality include the following: pneumonectomy (p < 0.001), bilobectomy (p < 0.001), American Society of Anesthesiology rating (p < 0.018), Zubrod performance status (p < 0.001), renal dysfunction (p = 0.001), induction chemoradiation therapy (p = 0.01), steroids (p = 0.002), age (p < 0.001), urgent procedures (p = 0.015), male gender (p = 0.013), forced expiratory volume in one second (p < 0.001), and body mass index (p = 0.015).

CONCLUSIONS

Thoracic surgeons participating in the STS GTDB perform lung cancer resections with a low mortality and morbidity. The risk-adjustment models created have excellent performance characteristics and identify important predictors of mortality and major morbidity for lung cancer resections. These models may be used to inform clinical decisions and to compare risk-adjusted outcomes for quality improvement purposes.