Estimations of treatment effects based on covariate adjusted nonparametric methods

Targeted estimation of heterogeneous treatment effect in observational survival analysis.

The importance of quality of life issues in health care practice and research is steadily growing. This growing interest fits into the definition of health as proposed by the World Health Organization (WHO) in 1948 (1). The WHO defines health as 'a state of complete physical, mental and social well-being and not merely the absence of disease and infirmity'. The attention to health-related quality of life is reflected in the increase in the use of quality-of-life evaluation as a technique of clinical research since 1973, when only five articles listed 'quality of life' as a reference key word in the Medline data base; during the subsequent five-year periods there were 195, 273, 490, and 1252 such articles (2). Also in the field of allergy it has been recognized that allergic disease comprise more than the classical signs and symptoms being part of physical disorders such as allergic rhinitis, asthma and the atopic eczema/dermatitis syndrome (AEDS) (3). In the last decades an increasing effort has been made to understand the socioeconomic burden of atopic disease in terms of effects on health-related quality of life (HRQL) and healthcare costs. It has been acknowledged in several consensus reports that rhinitis and asthma are associated with impairments in the patients' functioning in day-to-day life at home, at work and at school 4-8). With the introduction of questionnaires designed to measure asthma- 9-11) and rhinitis-associated impairments of quality of life (12) it is clear that patients may be bothered by sleep disorders, emotional problems, impairment in activities and social functioning. Also, in general terms, patients with asthma (13) and allergic rhinitis (14) are impaired in their physical and mental functioning, including vitality and the perception of general health. From daily medical practice it can be easily understood that AEDS has a major impact on HRQL. In a way, the use of questionnaires focused on skin disease 15-17) formally confirms this association. Quality of life, QOL, has divergent meanings for different people. Also, HRQL may be considered as ill-defined. More agreement has been reached about the four domains of QOL which are considered to be important: 1) physical status and functional abilities; 2) psychological status and well-being; 3) social functioning; 4) economic and/or vocational status and factors ( 18 ). As the true quality of life value cannot be measured directly, researchers and clinicians have to resort to series of questions (items) to measure this construct indirectly. Combinations of items yield scores referring to physical, mental and social domains. An HRQL instrument must meet several criteria. It should address each component (symptom, condition) that is important to the patient. Attributes of an instrument are described in Table 1. It will be clear that the construction of quality of life questionnaires is a complex task, drawing from the fields of clinimetrics, psychometrics and clinical decision-making (2). Differences in approach, for instance item selection using factor analysis vs the impact method which select items that are most frequently perceived as important by patients -- yields different questionnaires (19). In general two types of instruments, generic and specific, have been used in allergy research. Generic questionnaires measure physical, psychological and social domains in all health conditions irrespective of the underlying disease. A frequently used generic instrument is the Medical Outcomes Survey Short Form 36 (SF-36) (20). The SF-36 was developed as part of the Medical Outcomes Study and analyzes health status using 36 questions to measure nine different health dimensions. It has been used to characterize patients with asthma. Bousquet (13) compared the FEV1 and a clinical score of asthma severity for 252 asthmatic patients. There was a significant positive correlation between all nine quality of life domains of the SF-36 and the clinical score of Aas. Eight of the nine domains also correlated with the FEV1. Also in perennial rhinitis there was a significant impairment in eight of nine QOL dimensions in patients compared with healthy subjects (14). Furthermore, the SF-36 is used to evaluate the effects of a nonsedating antihistamine on quality of life. In this study all of the nine quality of life dimensions improved significantly after one and six weeks of cetirizine treatment compared with placebo (21). Other generic instruments that have been used in allergy research are the Sickness Impact Profile (SIP) (22) and the Nottingham Health Profile (NHP) (23). The 136 items in 12 categories of the SIP describe activities of everyday living. This instrument has been used to evaluate the effect of salmeterol on asthma (24). Salmeterol led to significant improvements over salbutamol on virtually all clinical outcomes. Although all four quality of life instruments used in this study showed the same trend in favor of salmeterol, only the disease-specific Asthma Quality of Life Questionnaire (AQLQ) and the Rating Scale utilities showed significantly greater improvement on salmeterol than on salbutamol. In severe AEDS it was shown, using the SIP, that cyclosporin improves quality of life significantly (25). In particular, the SIP has been used for comparison with disease-specific instruments (24, 26-28). The NHP, the only generic instrument derived entirely from lay people, has been used to validate a disease-specific instrument for patients with dermatitis and psoriasis (29). In asthma the NHP was not able to capture clinical improvement by treatment with pulmonary steroids (30). The latter observations underline the disadvantage that the generic instruments miss depth and therefore may not be responsive enough to detect changes in general health states in spite of important changes in disease-related problems (26). The advantage of generic instruments, however, is that the burden of illness across different disorders and patient populations can be compared. In a comparison between asthma and epilepsy the major finding was that children with epilepsy had a relatively more compromised quality of life in the psychological, social, and school domains (31. In contrast, children with asthma had a more compromised quality of life in the physical domain. These findings suggested that attention simply to seizure control in the clinical setting will not address the full range of quality of life problems in children with epilepsy. Specific instruments have been designed by asking patients what kind of problems they experience from their disease. Both the frequency and the importance of impairments are measured by means of the questionnaires. These instruments have the advantage that they describe the disease-associated problems of the patients. As stated above, they seem to be more responsive to changes in HRQL than do the generic instruments. Several instruments for patients with asthma have been developed. The Asthma Quality of Life Questionnaire of Juniper is focused on symptoms, emotions, exposure to environmental stimuli, and activity limitation (32). Modifications of this questionnaire have been published recently (33, 34). When using HRQL outcome in clinical trials, the question arises whether a change in HRQL is of clinical importance. For the AQLQ, which uses a seven-point scale, the minimal important difference of quality of life score per item is considered to be very close to 0.5 (35). A change of 1.0 in the score represents a moderate change and a change in score of greater than 2.0 represents a large change in HRQL. The minimal important difference as described by Juniper is based upon patient opinions. Measures such as the standardized response mean or the effect size can be used to standardize changes. These measures are based solely upon the distribution of the observed data, in particular upon the variance (36). Recently, it has been shown that both the SF-36 and AQLQ were able to characterize a group of patients with moderate asthma very well, whereas the AQLQ domains were found to have the best discriminative properties (37. The Asthma Quality of Life Questionnaire of Marks captures breathlessness, physical restrictions, mood disturbance and concerns for health (38). St. George's Respiratory Questionnaire (11) is designed for patients with asthma and chronic obstructive pulmonary disorder COPD. It can be applied in both reversible and fixed airway obstruction. In contrast to other questionnaires, the Living with Asthma Questionnaire (10) does not include impairments experienced as a direct consequence of asthma symptomatology. Other instruments are presented in Table 2. The properties of the most frequently used questionnaire are described in Table 3. Specific instruments have been developed for children and caregivers (Table 2). In addition, questionnaires have been constructed for different age-groups of patients with rhinitis (12, 39-41). A simple practical questionnaire technique for routine clinical use, the Dermatology Life Quality Index (DLQI) has been introduced to characterize patients with skin disorders (15). This instrument has been used to compare patients with psoriasis and dermatitis (42). Also versions for children are available: the Children's Dermatology Life Quality Index (CDLQI) and the Infant's Dermatology Life Quality Index (IDLQI) (16). Other questionnaires are the Skindex (43) the Dermatology-Specific Quality of Life (DSQL) (17) and the patient-generated Dermatology Quality of Life Scales (DQOLS) (44). Recently, a questionnaire has been developed to measure HRQL in patients with allergy to insect stings. Subsequently, this instrument has been used in the evaluation of venom immunotherapy (45). It appeared that venom immunotherapy resulted in a statistically and clinically significant improvement in HRQL. Both in clinical practice and in research physicians and investigators rely on physiological and objective measures, whenever possible. However in asthma an increase in FEV1 or a decrease in PC20 histamine or methacholine may occur without any improvement experienced by the patient. Medical intervention may improve physiologic measures, whereas for instance side-effects of drugs or the cumbersome aspects of subcutaneous immunotherapy may unfavorably influence day-to-day life and compliance with treatment. It has been put forward that the classical outcome variables may only partially characterize the disease of the patient. From that point of view it has been advocated to measure HRQL along with the conventional clinical indices (46). In line with this reasoning is the weak association between classical asthma measures and the outcome of HRQL questionnaires. Comparison between de AQLQ of Marks with asthma symptoms and lung function variables revealed that a change in AQLQ score was weakly correlated with change in symptom score (r = 0.37, 95% CI 0.04–0.64) and change in BHR (r = 0.38, 95% CI 0.06–0.64). The association with change in peak flow variability was weak (r = 0.12, 95% CI 0.26–0.47) (27). Similar observations have been reported by others 47-50). An interesting study shows that the mere presence of respiratory symptoms or a (gradually) reduced lung function is insufficient reason for patients to seek medical help. Subjects are more likely to consult their general practitioner once their quality of everyday life is affected or they experience variability in lung function (51). Also, rhinitis related quality of life appears to be moderately correlated to the more classical outcome variables used in clinical trials, such as daily symptom scores and nasal hyperreactivity (52). Another argument to use quality of life instruments lies in the headstart with respect to the knowledge of their validation, reliability and responsiveness compared to the common symptom scores or visual analogue scores (VAS) scales used at clinical trials. In the field of nasal allergy, validation or standardization of symptom scores has rarely been the subject of research. In asthma, even quite recently introduced measures, such as the number of symptom-free days, merit more attention in terms of standardization and validation (53). Other reasons to assess quality of life are conceivable. Measurement of quality of life can also be useful for screening purposes or for evaluation of therapy. Quality of life may be a determinant of effectiveness or efficacy of treatment. Moreover, its assessment might be relevant to striving for optimal decision-making. As the perception of patients is clearly important in the management of disease and patient compliance (Fig. 1), measurement of this 'dimension' by HRQL questionnaires in clinical trials may be justified. The emphasis on quality of life has sometimes resulted in a routine inclusion of HRQL questionnaires in clinical trials. The inclusion of such an instrument is valuable only if the changes can be interpreted by clinicians and contributes to optimal medical decision-making. In an editorial, criticism has been directed to the routine inclusion of such instruments when the structure of the evaluation and its rationale appears ill-defined (54). A model representing the relationships between clinical aspects of therapy, HRQL and factors influencing HRQL (adapted from Cramer and Spilker (17)). Generally in clinical trials the effect of treatment or intervention on HRQL runs parallel with the effect on conventional medical outcome measures. However, in some studies differences can be found. In a study evaluating the combined effect of steroids and antihistamines no differences were demonstrated between patients treated with antihistamine and steroids vs steroids alone in terms of quality of life, whereas for some patient-rated symptoms the combination turned out to be superior (55). In a large multicenter study comparing budesonide and fluticasone it was found that both drugs were equally effective in suppressing symptoms (56), although budesonide had a better effect on general quality of life (57). This might indicate that patients perceive differences not captured by conventional symptom scores. The reverse situation, i.e. significant effects on classical outcomes (symptom scores, medication use, peak flow or FEV1) without important change in two generic and two specific HRQL measures has been described in a study on the effect of formoterol, a long-acting α2-agonist, in mild to moderate asthmatic patients (58). The latter discrepancies can be explained by a limited performance of HRQL measures in mild asthmatic patients. Alternatively, it is possible that the minor changes in symptom scores and lung function due to the intervention are not perceived by patients as relevant. Moreover, patients with a chronic condition may adapt themselves to their disease. The strength of HRQL questionnaires, that is the patient-centred approach, is also one of its weaknesses. Perceptions of quality of life experienced by persons may shift in time. It is easy to understand that a dramatic personal accident or a serious disease will not only cause deterioration in quality of life but will eventually also influence the patient's values and internal standards. For instance, in a study of quality of life after radiotherapy for laryngeal cancer, a temporary deterioration of physical functioning and symptoms was reported, mostly caused by side-effects of treatment. Despite physical deterioration, there was an improvement of emotional functioning and mood after treatment, probably as a result of psychological adaptation and coping processes (59). It is possible also that in less dramatic circumstances, disease and treatments will induce shifts in perception due to changes in the patient's values. Such subjective changes in patients' perception are known as response shift. Socioeconomic status is an additional important independent factor influencing HRQL. In a recent study with asthmatic patients it was shown that socioeconomic status attributes to HRQL. More importantly, in this study it was difficult to separate out the unique effects of socioeconomic status and race/ethnicity (60). Recently, a significant relationship between the mental health of children with asthma and family functioning has been shown (61). These findings suggest that the domains comprising the HRQL of children with asthma are related to both disease and non-disease factors. Psychological functioning influences the burden of a specific disease. A study designed to assess the effects of depressive symptoms on asthma patients' reports of functional status and health-related quality of life revealed that asthma patients with more depressive symptoms reported worse health-related quality of life than asthma patients with similar disease activity, but fewer depressive symptoms (62). Interestingly, these findings were seen not only in generic (SF-36) but also in specific (AQLQ) instruments. This means that a disease-specific instrument may be also influenced by phenomena such as fear and depression. Finally, patients may either intentionally or unconsciously mask their symptoms or trivialize their diseases. They may tend to ignore or discount those problems which they believe are unrelated to their illness. Others may tend to give socially desirable answers. Response shifts and illusory mental health (63) are not easily captured with HRQL instruments, but they will certainly influence the outcome of a clinical trial, when HRQL is chosen as the primary endpoint. In summary, one has to realize that the translation of clinical effects of treatment into perceived and reported changes in quality of life finds a place at the integration level of the patient and this is, in a way, a black box which is not easy to assess (Fig. 1). For these reasons it is strongly recommended to use HRQL outcome measures in parallel with conventional physiological outcome measures. Asthma, allergic rhinitis and AEDS often coexist. The question to what extent concomitant allergic disease affects quality of life has infrequently been addressed. In a recent study the SF-36 questionnaire from 850 subjects recruited in two French centers participating in the European Community Respiratory Health Survey was evaluated. Both asthma and allergic rhinitis were associated with impairment in quality of life. However, 78% of asthmatics also had allergic rhinitis. Subjects with allergic rhinitis but not asthma were more likely to report problems with social activities, difficulties with daily activities as a result of emotional problems, and low mental well-being than subjects with neither asthma nor rhinitis. Patients with both asthma and allergic rhinitis experienced more physical limitations than patients with allergic rhinitis alone, but no difference was found between these two groups for concepts related to social/mental health (64). In another study focusing on asthma, rhinitis and AEDS, comprising 325 subjects allergic to house dust mites, it was found that patients did show impaired quality of life compared to irrespective of the of the atopic Patients with the of asthma did out in terms of physical In addition, asthma symptoms with a visual had a major effect on social functioning, emotional functioning and disorders, in patients with AEDS, appeared to be associated with physical functioning, social functioning, mental health and general health It is not only concomitant atopic disease that has an impact on quality of life. such as and and nasal may patients with rhinitis and asthma. the SF-36 and a quality of life measure it has been shown that HRQL is impaired and that may improve quality of life for patients that is a other specific instruments such as the Index and the have been The impact of on social life in children during the four of life is not easily can be by use of a specific which measures the quality of life is a chronic disease of the respiratory which is frequently associated with respiratory compared the HRQL in patients with nasal with those of patients with perennial rhinitis and healthy It appeared that nasal impaired HRQL more than perennial allergic rhinitis The impairment of HRQL was greater when nasal was associated with asthma In addition, of nasal symptoms, and pulmonary function were after the evaluation in patients with nasal These demonstrated that nasal treatment either with nasal steroids or significantly improved both nasal symptoms and QOL without significant changes in pulmonary may a if the or is in one particular disease. A recent study the effects of on the of QOL measures an analysis of data from clinical trials with asthma, and The study suggest that conditions significantly and patients' scores on generic QOL measures and of treatment whereas their influence on disease-specific QOL scores and of treatment effect is although not These findings have significant practical for the of true treatment control of and the of QOL trials. The that atopic disease may have an effect on daily functioning has been by studies focused on school and in children with asthma may school and as as work by In a study it was shown that of children with recent symptoms of asthma, reported school absence for at one during the 12 compared with in children without respiratory absence of respiratory illness was reported for and use for respiratory problems for of the children with recent symptoms of asthma In another study reported in their activities and reported of work and school of asthma or nasal symptoms are not in patients with allergic rhinitis they may to problems during school either by direct or of sleep and allergic rhinitis may be associated with reduced to with will these problems, whereas treatment with nonsedating will only partially reverse the limitations in Recently, in a study out over in children with allergic perennial rhinitis and children with perennial rhinitis, it was shown that or the from on school on school and sleep In of the of a large it has been demonstrated that in asthma with increasing disease severity The of the effect of asthma on work the effect of work on asthma. The of asthma and of asthma is increasing It has been that of asthma can be to of asthma at work more on the of of underlying asthma than on the of possible asthma. It can be that patients with asthma may have a more severe impairment in quality of life of the between work and disease. In a study designed to address this question a statistically significant difference was seen in the scores of the AQLQ from a group of patients with asthma and a control group of subjects with asthma of The mean difference in the score was on a of limitation or of the to limitation or all the at the of the patient with asthma The difference between both groups was other more generic instruments focused on detect more showed that both asthma and rhinitis work with asthma are less likely to be at those rhinitis is a more determinant of work effectiveness In the allergic rhinitis in school days, and reduced activity per These data are derived from persons allergic rhinitis in with persons medical treatment. These data indicate that allergic rhinitis may have an important impact on and Patients are bothered by with performance and at and and may and only disease but also may influence work It has been that of treated their allergic rhinitis with antihistamines at for per Patients these antihistamines are more likely to The of include and With the antihistamines these problems have been significantly reduced studies have the for treatment of allergic rhinitis, asthma and associated In asthma in the for an A comparison of asthma in developed suggested an burden from to per of the asthma were to direct medical For the it has been that the when allergic was the primary were in The when allergic was a to other disorders such as asthma and was at The of allergic asthma and rhinitis and concerns about health care the increasing interest for only does the efficacy of treatment have to be but also its In these studies measures must be in to across patient populations and for different It is, however, difficult to the generic SF-36 or disease-specific HRQL scores into For this utilities such as the have been which measure the value that patients themselves place on their health some utilities measure the value that on health are the and Health An advantage of utilities is their to life associated with different medical can easily be into instruments are mostly A recent rhinitis specific the has been developed as a patient outcome for clinical trials and for studies comparing medical treatments for rhinitis The same group introduced an asthma specific the Asthma Index Also, disease-specific versions of the and have been developed for patients with asthma The interest in quality of life for patients with allergy that allergy is by a significant socioeconomic the introduction of HRQL outcome measures physicians were that patients cannot be by physiological measures. In a way, HRQL outcome measures of the from the with which clinicians are in their day-to-day The of these in the HRQL questionnaires it possible to include the patient in clinical trials and the in this field will improve medical decision-making and management of disease. of these outcome measures in the evaluation and management of patients be the However, HRQL questionnaires are in the of being in terms of and introduction of of instruments of QOL data is based on the that there are no measurement in the of is an technique for and which measurement into An important of is that it of whether a model fits the observed With this it has been shown that some changes in the of the SF-36 are when it is applied to evaluation of QOL for patients with or disease and with experienced criticism has been the of instruments and the to the measurement of quality of life It has been that attention has to be to better for of and of measures, these instruments will be for use in clinical practice and for use as primary in clinical trials Also, in the field of allergy the number of outcome measures is growing. For the and it will be difficult to select the of questionnaires. A a clinical is in of an disease-specific questionnaire with a whereas a at the level of health a generic instrument differences between subjects at a point in and utilities to assess of In not to patients with outcome measures research is to between In research to be focused on the selection and of a limited number of and instruments in to better understand the patient with allergy and better the of clinical trials. from the of Medical and for of the and

Effects of comorbidity on health-related quality-of-life scores: an analysis of clinical trial data

Causal inferences on the average treatment effect in observational studies are always difficult problems because the distributions of samples in the two treatment groups can not be observed at the same time, and the estimation of the treatment effect is often biased.In this paper, the propensity score and the propensity score subclassification, selected from several methods, are used to assess the treatment effect.The estimation of the average treatment effect give the Bootstrap confidence intervals. Simulation studies are inducted for the continuous samples in normal distribution and the mixed samples of discrete and continuous type.

Survey satisficing in online data collection biases the estimation of treatment effects in many ways. Extending the findings of a previous study, which demonstrated that satisficing biased the estimation of main treatment effects, this study also shows that satisficing distorts the estimation of moderation effects. Targeting Japanese adults’ attitudes toward food, this study tests how the effect of country of production (Japan vs. China) is moderated by preexisting ethnocentric attitudes. The results show that while nonsatisficers predictably adjust their attitude toward food based on their preexisting ethnocentric attitude, satisficers stick to their initial stereotypical response. That is, the theoretically predicted moderation effect was observed among nonsatisficers, but not among satisficers, which indicates that satisficing biases not only the estimation of a treatment effect but also that of a moderation effect.

IntroductionThe real-world treatment effect of a novel treatment can be estimated by analysing routinely collected patient data, in the form of Electronic Health Records (EHR). Any treatment allocation in EHR is not randomised and there may be systematic differences between the treatment groups. Propensity Score (PS) methods are commonly used to correct for these differences and reduce the bias in the treatment effect estimate. The aims of the study were to compare the performance of the most popular PS methods in the estimation of the treatment effect in the presence of two common issues in EHRs: covariate measurement error and sparse data.MethodsThe motivational example for this study was the assessment of the treatment effect of the novel oral anti-coagulant Rivaroxaban compared with the previous standard treatment Warfarin for the prevention of future stroke in patients with atrial fibrillation. Using simulation experiments based on a dataset comparing Rivaroxaban with Warfarin, we evaluated the performance of four PS methods.ResultsIn the simulations with characteristics of the original dataset, using 3:1 PS matching generated a largest bias of +0.0428 (corresponding ratio of HRs (rHR) 1.0437), whereas for the other PS methods it was smaller and in negative direction: IPTW for ATE -0.0181 (rHR = 0.9821); IPTW for ATT -0.0110 (rHR = 0.9891); PS stratification −0.0099 (rHR = 0.9901), with relative differences between rHRs being small to negligible. Fifty percent under-recording of a covariate (stroke) in the PS model, increased the MSE between 6% and 11% compared to the MSE with no introduced measurement error. While 50% over-recording reduced the MSE by around 35%. The difference in the bias of the low prevalence outcome (0.5%) and the high prevalence outcome (10%) was: IPTW for ATE 0.1514 (rHRs = 1.1635); IPTW for ATT 0.0160 (rHRs = 1.0161); 3:1 PS matching 0.0758 (rHRs = 1.0787); PS Stratification 0.0177 (rHRs = 1.0179). A similar pattern for outcome prevalence was seen for all the simulation scenarios.ConclusionThis study showed that PS methods proposed in the literature may not all perform well for individual datasets. The findings produced recommendations for using PS methods in the estimation of real-world treatment effect when the covariate measurement error and sparse outcome data are present.

Letter to the Editor Re: Orthopaedic Surgeons Prefer to Participate in Expertise-based Randomized Trials

Estimation of treatment effects with causal interpretation from observational data is complicated because exposure to treatment may be confounded with subject characteristics. The propensity score, the probability of treatment exposure conditional on covariates, is the basis for two approaches to adjusting for confounding: methods based on stratification of observations by quantiles of estimated propensity scores and methods based on weighting observations by the inverse of estimated propensity scores. We review popular versions of these approaches and related methods offering improved precision, describe theoretical properties and highlight their implications for practice, and present extensive comparisons of performance that provide guidance for practical use.

Estimation of treatment effects with causal interpretation from observational data is complicated because exposure to treatment may be confounded with subject characteristics. The propensity score, the probability of treatment exposure conditional on covariates, is the basis for two approaches to adjusting for confounding: methods based on stratification of observations by quantiles of estimated propensity scores and methods based on weighting observations by the inverse of estimated propensity scores. We review popular versions of these approaches and related methods offering improved precision, describe theoretical properties and highlight their implications for practice, and present extensive comparisons of performance that provide guidance for practical use.

In this article, I present a counterfactual model identifying average treatment effects by conditional mean independence when considering peer- or neighborhood-correlated effects, and I provide a new command, ntreatreg, that implements such models in practical applications. The model and its accompanying command provide an estimation of average treatment effects when the stable unit treatment-value assumption is relaxed under specific conditions. I present two instructional applications: the first is a simulation exercise that shows both model implementation and ntreatreg correctness; the second is an application to real data, aimed at measuring the effect of housing location on crime in the presence of social interactions. In the second application, results are compared with a no-interaction setting.

In this article, I present a counterfactual model identifying average treatment effects by conditional mean independence when considering peer- or neighborhood-correlated effects, and I provide a new command, ntreatreg, that implements such models in practical applications. The model and its accompanying command provide an estimation of average treatment effects when the stable unit treatment-value assumption is relaxed under specific conditions. I present two instructional applications: the first is a simulation exercise that shows both model implementation and ntreatreg correctness; the second is an application to real data, aimed at measuring the effect of housing location on crime in the presence of social interactions. In the second application, results are compared with a no-interaction setting.

We present the L2Boosting algorithm and two variants, namely post-Boosting and orthogonal Boosting. Building on results in Ye and Spindler (2016), we demonstrate how boosting can be used for estimation and inference of low-dimensional treatment effects. In particular, we consider estimation of a treatment effect in a setting with very many controls and in a setting with very many instruments. We provide simulations and analyze two real applications. We compare the results with Lasso and find that boosting performs quite well. This encourages further use of boosting for estimation of treatment effects in high-dimensional settings.

In this paper, the role of the propensity score in the efficient estimation of average treatment effects is examined. Under the assumption that the treatment is ignorable given some observed characteristics, it is shown that the propensity score is ancillary for estimation of the average treatment effects. The propensity score is not ancillary for estimation of average treatment effects on the treated. It is suggested that the marginal value of the propensity score lies entirely in the dimension reduction. Efficient semiparametric estimators of average treatment effects and average treatment effects on the treated are shown to take the form of relevant sample averages of the data completed by the nonparametric imputation method. It is shown that the projection on the propensity score is not necessary for efficient semiparametric estimation of average treatment effects on the treated even if the propensity score is known. An application to the experimental data reveals that conditioning on the propensity score may even result in a loss of efficiency.

Fibrotic interstitial lung diseases (ILDs) cause loss of forced vital capacity (FVC) and increased risk of death over time. Most clinical trials aim to slow FVC decline and reduce mortality. However, the association of lower FVC with higher mortality will bias simple estimates of differences in FVC progression between groups. Therefore, both the time-dependent decline in FVC and increase in mortality should be jointly modeled. We developed a Bayesian, joint mixed-effects disease progression model (DPM), using minimally informative prior distributions, for FVC trajectory and the hazard for ILD-related mortality over time. This model minimizes bias due to mortality in estimating differences in the rate of FVC decline and is suitable for use when characterizing populations or in estimating a treatment effect in a clinical trial. The DPM was applied to individual patient data from prospective cohort studies of fibrotic ILD. The DPM yields a higher estimated rate of FVC decline (6.0% vs 4.7%/year) and a more precise fit than a linear mixed model of FVC alone, and replicates the nonlinear pattern in the observed data. By modeling the full FVC trajectory rather than only the change from baseline at a given time point, the DPM increases the information from each patient and reduces both the time to information and the effect of variability in baseline FVC measurements on the estimation of treatment effects. The joint DPM provides an integrated approach to minimizing bias in the estimation of treatment effects in clinical trials in fibrotic ILDs.

SummaryThe estimation of treatment effects based on observational data usually involves multiple confounders, and dimension reduction is often desirable and sometimes inevitable. We first clarify the definition of a central subspace that is relevant for the efficient estimation of average treatment effects. A criterion is then proposed to simultaneously estimate the structural dimension, the basis matrix of the joint central subspace, and the optimal bandwidth for estimating the conditional treatment effects. The method can easily be implemented by forward selection. Semiparametric efficient estimation of average treatment effects can be achieved by averaging the conditional treatment effects with a different data-adaptive bandwidth to ensure optimal undersmoothing. Asymptotic properties of the estimated joint central subspace and the corresponding estimator of average treatment effects are studied. The proposed methods are applied to a nutritional study, where the covariate dimension is reduced from 11 to an effective dimension of one.