Faculty Bibliography

In causal inference, a variety of causal effect estimands have been studied, including the sample, uncensored, target, conditional, optimal subpopulation, and optimal weighted average treatment effects. Ad hoc methods have been developed for each estimand based on inverse probability weighting (IPW) and on outcome regression modeling, but these may be sensitive to model misspecification, practical violations of positivity, or both. The contribution of this article is twofold. First, we formulate the generalized average treatment effect (GATE) to unify these causal estimands as well as their IPW estimates. Second, we develop a method based on Kernel optimal matching (KOM) to optimally estimate GATE and to find the GATE most easily estimable by KOM, which we term the Kernel optimal weighted average treatment effect. KOM provides uniform control on the conditional mean squared error of a weighted estimator over a class of models while simultaneously controlling for precision. We study its theoretical properties and evaluate its comparative performance in a simulation study. We illustrate the use of KOM for GATE estimation in two case studies: comparing spine surgical interventions and studying the effect of peer support on people living with HIV.

Age is an important patient characteristic that has been correlated with specific outcomes after lumbar spine surgery. We performed a retrospective cohort study to model the effect of age on discharge destination and complications after a 1-level or multi-level lumbar spine fusion surgery. The American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database was used to identify patients who underwent lumbar spinal fusion surgery from 2013 through 2017. Perioperative outcomes were compared across ages 18 to 90 using multivariable nonlinear logistic regressioncontrolling for preoperative characteristics. A total of 61,315 patients were analyzed, with patients over 70 having a higher risk of being discharged to an inpatient rehabilitation center and receiving an intraoperative or postoperative blood transfusion. However, the rates of the other complications and outcomes analyzed in this study were not significantly different as patients age. In conclusion, advanced-age affects the discharge destination after a one- or multi-level fusion and intraoperative/postoperative blood transfusion after a one-level fusion. However, age alone does not significantly affect the risk of the other complications and outcomes assessed in this study. This study will help guide preoperative discussion with advanced-aged patients who are considering a 1-level or multi-level lumbar spine fusion surgery.

Inverse probability of treatment weighting (IPTW), which has been used to estimate average treatment effects (ATE) using observational data, tenuously relies on the positivity assumption and the correct specification of the treatment assignment model, both of which are problematic assumptions in many observational studies. Various methods have been proposed to overcome these challenges, including truncation, covariate-balancing propensity scores, and stable balancing weights. Motivated by an observational study in spine surgery, in which positivity is violated and the true treatment assignment model is unknown, we present the use of optimal balancing by kernel optimal matching (KOM) to estimate ATE. By uniformly controlling the conditional mean squared error of a weighted estimator over a class of models, KOM simultaneously mitigates issues of possible misspecification of the treatment assignment model and is able to handle practical violations of the positivity assumption, as shown in our simulation study. Using data from a clinical registry, we apply KOM to compare two spine surgical interventions and demonstrate how the result matches the conclusions of clinical trials that IPTW estimates spuriously refute.

Marginal structural models (MSMs) can be used to estimate the causal effect of a potentially time-varying treatment in the presence of time-dependent confounding via weighted regression. The standard approach of using inverse probability of treatment weighting (IPTW) can be sensitive to model misspecification and lead to high-variance estimates due to extreme weights. Various methods have been proposed to partially address this, including covariate balancing propensity score (CBPS) to mitigate treatment model misspecification, and truncation and stabilized-IPTW (sIPTW) to temper extreme weights. In this article, we present kernel optimal weighting (KOW), a convex-optimization-based approach that finds weights for fitting the MSMs that flexibly balance time-dependent confounders while simultaneously penalizing extreme weights, directly addressing the above limitations. We further extend KOW to control for informative censoring. We evaluate the performance of KOW in a simulation study, comparing it with IPTW, sIPTW, and CBPS. We demonstrate the use of KOW in studying the effect of treatment initiation on time-to-death among people living with human immunodeficiency virus and the effect of negative advertising on elections in the United States.

OBJECTIVE:Spine surgery is especially susceptible to malpractice claims. Critics of the US medical liability system argue that it drives up costs, whereas proponents argue it deters negligence. Here, the authors study the relationship between malpractice claim density and outcomes. METHODS:The following methods were used: 1) the National Practitioner Data Bank was used to determine the number of malpractice claims per 100 physicians, by state, between 2005 and 2010; 2) the Nationwide Inpatient Sample was queried for spinal fusion patients; and 3) the Area Resource File was queried to determine the density of physicians, by state. States were categorized into 4 quartiles regarding the frequency of malpractice claims per 100 physicians. To evaluate the association between malpractice claims and death, discharge disposition, length of stay (LOS), and total costs, an inverse-probability-weighted regression-adjustment estimator was used. The authors controlled for patient and hospital characteristics. Covariates were used to train machine learning models to predict death, discharge disposition not to home, LOS, and total costs. RESULTS:Overall, 549,775 discharges following spinal fusions were identified, with 495,640 yielding state-level information about medical malpractice claim frequency per 100 physicians. Of these, 124,425 (25.1%), 132,613 (26.8%), 130,929 (26.4%), and 107,673 (21.7%) were from the lowest, second-lowest, second-highest, and highest quartile states, respectively, for malpractice claims per 100 physicians. Compared to the states with the fewest claims (lowest quartile), surgeries in states with the most claims (highest quartile) showed a statistically significantly higher odds of a nonhome discharge (OR 1.169, 95% CI 1.139-1.200), longer LOS (mean difference 0.304, 95% CI 0.256-0.352), and higher total charges (mean difference [log scale] 0.288, 95% CI 0.281-0.295) with no significant associations for mortality. For the machine learning models-which included medical malpractice claim density as a covariate-the areas under the curve for death and discharge disposition were 0.94 and 0.87, and the R2 values for LOS and total charge were 0.55 and 0.60, respectively. CONCLUSIONS:Spinal fusion procedures from states with a higher frequency of malpractice claims were associated with an increased odds of nonhome discharge, longer LOS, and higher total charges. This suggests that medicolegal climate may potentially alter practice patterns for a given spine surgeon and may have important implications for medical liability reform. Machine learning models that included medical malpractice claim density as a feature were satisfactory in prediction and may be helpful for patients, surgeons, hospitals, and payers.

Marginal structural Cox models have been used to estimate the causal effect of a time-varying treatment on a survival outcome in the presence of time-dependent confounders. These methods rely on the positivity assumption, which states that the propensity scores are bounded away from zero and one. Practical violations of this assumption are common in longitudinal studies, resulting in extreme weights that may yield erroneous inferences. Truncation, which consists of replacing outlying weights with less extreme ones, is the most common approach to control for extreme weights to date. While truncation reduces the variability in the weights and the consequent sampling variability of the estimator, it can also introduce bias. Instead of truncated weights, we propose using optimal probability weights, defined as those that have a specified variance and the smallest Euclidean distance from the original, untruncated weights. The set of optimal weights is obtained by solving a constrained quadratic optimization problem. The proposed weights are evaluated in a simulation study and applied to the assessment of the effect of treatment on time to death among people in Sweden who live with human immunodeficiency virus and inject drugs.

BACKGROUND:People who inject drugs (PWID) frequently engage in injection risk behaviours exposing them to blood-borne infections. Understanding the underlying causes that drive various types and levels of risk behaviours is important to better target preventive interventions. METHODS:A total of 2150 PWID in Swedish remand prisons were interviewed between 2002 and 2012. Questions on socio-demographic and drug-related variables were asked in relation to the following outcomes: Having shared injection drug solution and having lent out or having received already used drug injection equipment within a 12 month recall period. RESULTS:Women shared solutions more than men (odds ratio (OR) 1.51, 95% confidence interval (CI) 1.03; 2.21). Those who had begun to inject drugs before age 17 had a higher risk (OR 1.43, 95% CI 0.99; 2.08) of having received used equipment compared to 17-19 year olds. Amphetamine-injectors shared solutions more than those injecting heroin (OR 2.43, 95% CI 1.64; 3.62). A housing contract lowered the risk of unsafe injection by 37-59% compared to being homeless. CONCLUSIONS:Women, early drug debut, amphetamine users and homeless people had a significantly higher level of injection risk behaviour and need special attention and tailored prevention to successfully combat hepatitis C and HIV transmission among PWID. TRIAL REGISTRATION:ClinicalTrials.gov Identifier, NCT02234167.

BACKGROUND:Switch from first line antiretroviral therapy (ART) to second-line ART is common in clinical practice. However, there is limited knowledge of to which extent different reason for therapy switch are associated with differences in long-term consequences and sustainability of the second line ART. MATERIAL AND METHODS/METHODS:Data from 869 patients with 14601 clinical visits between 1999-2014 were derived from the national cohort database. Reason for therapy switch and viral load (VL) levels at first-line ART failure were compared with regard to outcome of second line ART. Using the Laplace regression model we analyzed the median, 10th, 20th, 30th and 40th percentile of time to viral failure (VF). RESULTS:Most patients (n = 495; 57.0%) switched from first-line to second-line ART without VF. Patients switching due to detectable VL with (n = 124; 14.2%) or without drug resistance mutations (DRM) (n = 250; 28.8%) experienced VF to their second line regimen sooner (median time, years: 3.43 (95% CI 2.90-3.96) and 3.20 (95% 2.65-3.75), respectively) compared with those who switched without VF (4.53 years). Furthermore level of VL at first-line ART failure had a significant impact on failure of second-line ART starting after 2.5 years of second-line ART. CONCLUSIONS:In the context of life-long therapy, a median time on second line ART of 4.53 years for these patients is short. To prolong time on second-line ART, further studies are needed on the reasons for therapy changes. Additionally patients with a high VL at first-line VF should be more frequently monitored the period after the therapy switch.

BACKGROUND:The effect of peer support on virologic and immunologic treatment outcomes among HIVinfected patients receiving antiretroviral therapy (ART) was assessed in a cluster randomized controlled trial in Vietnam. METHODS:Seventy-one clusters (communes) were randomized in intervention or control, and a total of 640 patients initiating ART were enrolled. The intervention group received peer support with weekly home-visits. Both groups received first-line ART regimens according to the National Treatment Guidelines. Viral load (VL) (ExaVir™ Load) and CD4 counts were analyzed every 6 months. The primary endpoint was virologic failure (VL >1000 copies/ml). Patients were followed up for 24 months. Intention-to-treat analysis was used. Cluster longitudinal and survival analyses were used to study time to virologic failure and CD4 trends. RESULTS:Of 640 patients, 71% were males, mean age 32 years, 83% started with stavudine/lamivudine/nevirapine regimen. After a mean of 20.8 months, 78% completed the study, and the median CD4 increase was 286 cells/μl. Cumulative virologic failure risk was 7.2%. There was no significant difference between intervention and control groups in risk for and time to virologic failure and in CD4 trends. Risk factors for virologic failure were ART-non-naïve status [aHR 6.9;(95% CI 3.2-14.6); p < 0.01]; baseline VL ≥100,000 copies/ml [aHR 2.3;(95% CI 1.2-4.3); p < 0.05] and incomplete adherence (self-reported missing more than one dose during 24 months) [aHR 3.1;(95% CI 1.1-8.9); p < 0.05]. Risk factors associated with slower increase of CD4 counts were: baseline VL ≥100,000 copies/ml [adj.sq.Coeff (95% CI): -0.9 (-1.5;-0.3); p < 0.01] and baseline CD4 count <100 cells/μl [adj.sq.Coeff (95% CI): -5.7 (-6.3;-5.4); p < 0.01]. Having an HIV-infected family member was also significantly associated with gain in CD4 counts [adj.sq.Coeff (95% CI): 1.3 (0.8;1.9); p < 0.01]. CONCLUSION:There was a low virologic failure risk during the first 2 years of ART follow-up in a rural low-income setting in Vietnam. Peer support did not show any impact on virologic and immunologic outcomes after 2 years of follow up. TRIAL REGISTRATION:NCT01433601 .