Covariate Balance Research Articles

Reduced human fecundity attributable to ambient fine particles in low- and middle-income countries

BackgroundExposure to ambient fine particulate matter (PM2.5) has been associated with reduced human fecundity. However, the attributable burden has not been estimated for low- and middle-income countries (LMICs), where the exposure–response function between PM2.5 and the infertility rate has been insufficiently studied. ObjectiveThis study examined the associations between long-term exposure to PM2.5 and human fecundity indicators, namely the expected time to pregnancy (TTP) and 12-month infertility rate (IR), and then estimated PM2.5-attributable burden of infertility in LMICs. MethodsWe analyzed 164,593 eligible women from 100 Demographic and Health Surveys conducted in 49 LMICs between 1999 and 2021. We assessed PM2.5 exposures during the 12 months before a pregnancy attempt using the global satellite-derived PM2.5 estimates produced by Atmospheric Composition Analysis Group (ACAG). First, we created a series of pseudo-populations with balanced covariates, given different levels of PM2.5 exposure, using a matching approach based on the generalized propensity score. For each pseudo-population, we used 2-stage generalized Gamma models to derive TTP or IR from the probability distribution of the questionnaire-based duration time for the pregnancy attempt before the interview. Second, we used spline regressions to generate nonlinear PM2.5 exposure–response functions for each of the two fecundity indicators. Finally, we applied the exposure–response functions to estimate number of infertile couples attributable to PM2.5 exposure in 118 LMICs. ResultsBased on the Gamma models, each 10 µg/m3 increment in PM2.5 exposure was associated with a TTP increase by 1.7 % (95 % confidence interval [CI]: -2.3 %–6.0 %) and an IR increase by 2.3 % (95 %CI: 0.6 %–3.9 %). The nonlinear exposure–response function suggested a robust effect of an increased IR for high-concentration PM2.5 exposure (>75 µg/m3). Based on the PM2.5-IR function, across the 118 LMICs, the number of infertile couples attributable to PM2.5 exposure exceeding 35 µg/m3 (the first-stage interim target recommended by the World Health Organization global air quality guidelines) was 0.66 million (95 %CI: 0.061–1.43), accounting for 2.25 % (95 %CI: 0.20 %–4.84 %) of all couples affected by infertility. Among the 0.66 million, 66.5 % were within the top 10 % high-exposure infertile couples, mainly from South Asia, East Asia, and West Africa. ConclusionPM2.5 contributes significantly to human infertility in places with high levels of air pollution. PM2.5-pollution control is imperative to protect human fecundity in LMICs.

Balancing versus modelling in weighted analysis of non-randomised studies with survival outcomes: A simulation study.

Weighting methods are widely used for causal effect estimation in non-randomised studies. In general, these methods use the propensity score (PS), the probability of receiving the treatment given the covariates, to arrive at the respective weights. All of these "modelling" methods actually optimize prediction of the respective outcome, which is, in the PS model, treatment assignment. However, this does not match with the actual aim of weighting, which is eliminating the association between covariates and treatment assignment. In the "balancing" approach, covariates are thus balanced directly by solving systems of numerical equations, explicitly without fitting a PS model. To compare modelling, balancing and hybrid approaches to weighting we performed a large simulation study for a binary treatment and a survival outcome. For maximal practical relevance all simulation parameters were selected after a systematic review of medical studies that used PS methods for analysis. We also introduce a new hybrid method that uses the idea of the covariate balancing propensity score and matching weights, thus avoiding extreme weights. In addition, we present a corrected robust variance estimator for some of the methods. Overall, our simulations results indicate that balancing approach methods work worse than expected. However, among the considered balancing methods, entropy balancing consistently outperforms the variance balancing approach. All methods estimating the average treatment effect in the overlap population perform well with very little bias and small standard errors even in settings with misspecified propensity score models. Finally, the coverage using the standard robust variance estimator was too high for all methods, with the proposed corrected robust variance estimator improving coverage in a variety of settings.

Pigeonhole Design: Balancing Sequential Experiments from an Online Matching Perspective

Practitioners and academics have long appreciated the benefits of covariate balancing when they conduct randomized experiments. For web-facing firms running online A/B tests, however, it still remains challenging in balancing covariate information when experimental subjects arrive sequentially. In this paper, we study an online experimental design problem, which we refer to as the online blocking problem. In this problem, experimental subjects with heterogeneous covariate information arrive sequentially and must be immediately assigned into either the control or the treated group. The objective is to minimize the total discrepancy, which is defined as the minimum weight perfect matching between the two groups. To solve this problem, we propose a randomized design of experiment, which we refer to as the pigeonhole design. The pigeonhole design first partitions the covariate space into smaller spaces, which we refer to as pigeonholes, and then, when the experimental subjects arrive at each pigeonhole, balances the number of control and treated subjects for each pigeonhole. We analyze the theoretical performance of the pigeonhole design and show its effectiveness by comparing against two well-known benchmark designs: the matched-pair design and the completely randomized design. We identify scenarios when the pigeonhole design demonstrates more benefits over the benchmark design. To conclude, we conduct extensive simulations using Yahoo! data to show a 10.2% reduction in variance if we use the pigeonhole design to estimate the average treatment effect. This paper was accepted by George Shanthikumar, data science. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.02184 .

Emergency Department Referral of Patients With Chest Pain for Noninvasive Cardiac Testing and 2-Year Clinical Outcomes.

Noninvasive cardiac testing (NICT) has been associated with decreased long-term risks of major adverse cardiac events (MACEs) among emergency department patients at high coronary risk. It is unclear whether this association extends to patients without evidence of myocardial injury on initial ECG and cardiac troponin testing. A retrospective cohort study was conducted of patients presenting with chest pain between 2013 and 2019 to 21 emergency departments within an integrated health care system in Northern California, excluding patients with ST-segment-elevation myocardial infarction or myocardial injury by serum troponin testing. To account for confounding by indication, we grouped patient encounters by the NICT referral rate of the initially assigned emergency physician relative to local peers within discrete time periods. The primary outcome was MACE within 2 years. Secondary outcomes were coronary revascularization and MACE, inclusive of all-cause mortality. Associations between the NICT referral group (low, intermediate, or high) and outcomes were assessed using risk-adjusted proportional hazards methods with censoring for competing events. Among 144 577 eligible patient encounters, the median age was 58 years (interquartile range, 48-68) and 57% were female. Thirty-day NICT referral was 13.0%, 19.9%, and 27.8% in low, intermediate, and high NICT referral groups, respectively, with a good balance of baseline covariates between groups. Compared with the low NICT referral group, there was no significant decrease in the adjusted hazard ratio of MACE within the intermediate (adjusted hazard ratio, 1.08 [95% CI, 1.02-1.14]) or high (adjusted hazard ratio, 1.05 [95% CI, 0.99-1.11]) NICT referral groups. Results were similar for MACE, inclusive of all-cause mortality, and coronary revascularization, as well as subgroup analyses stratified by estimated risk (history, electrocardiogram, age, risk factors, troponin [HEART] score: percent classified as low risk, 48.2%; moderate risk, 49.2%; and high risk, 2.7%). Increases in NICT referrals were not associated with changes in the hazard of MACE within 2 years following emergency department visits for chest pain without evidence of acute myocardial injury. These findings further highlight the need for evidence-based guidance regarding the appropriate use of NICT in this population.

Textbook oncologic outcomes among patients undergoing laparoscopic, robotic and open surgery for intrahepatic and perihilar cholangiocarcinoma

BackgroundTextbook oncologic outcome (TOO) serves as a composite, oncologic metric for surgical quality of care. We sought to evaluate variations in TOO among patients undergoing laparoscopic, robotic, and open surgery for intrahepatic (iCCA) and perihilar (pCCA) cholangiocarcinoma. MethodsPatients who underwent liver resection for iCCA and pCCA between 2010 and 2018 were identified from the National Cancer Database. Entropy balancing was performed for covariate balancing and multivariable regression was used to evaluate the association between surgical approach and TOO. ResultsAmong 5434 patients who underwent hepatic resection between 2010 and 2018, 3888 (71.6%) had iCCA, and 1546 (28.4%) had pCCA. TOO was achieved in 11.7% (n = 454), and 18.8% (n = 291) of patients with iCCA and pCCA, respectively. There was a difference in achievement of TOO relative to operative approach among patients with iCCA (robotic: 6.2% vs. laparoscopic: 8.1% vs. open: 12.5%; p = 0.002). After entropy balancing, patients with iCCA undergoing laparoscopic surgery had 32% reduced odds of achieving TOO (Ref: open surgery; laparoscopic, OR 0.68, 95%CI 0.49–0.93; p = 0.016; robotic, OR 0.69, 95%CI 0.34–1.39; p = 0.298). ConclusionsUsage of composite oncologic measures such as TOO may allow for a holistic assessment of different approaches to hepatic resection among patients with CCA.

Preoperative Risk Assessment of Early Kidney Graft Loss.

A large proportion of potential organ donors are not utilized for kidney transplantation out of risk of early allograft loss because of donor-related characteristics. These can be summarized using kidney donor profile index (KDPI). Because KDPI affects the choice of the recipient, the predictive ability of KDPI is tied to recipient attributes. These have been questioned to explain most of the predictive ability of KDPI. This study aims to quantify the effect of the donor on early graft loss (EGL) by accounting for nonrandom allocation. This study included patients undergoing kidney transplantation from deceased donors between 2014 and 2020 from the Scientific Registry of Transplantation Recipients. EGL, defined as a return to dialysis or retransplantation during the first posttransplant year, was the primary endpoint. Nonrandom allocation and donor-recipient matching by KDPI necessitated the use of inverse probability treatment weighting, which served to assess the effect of KDPI and mitigate selection bias in a weighted Cox regression model. The study comprised 89 290 transplantations in 88 720 individual patients. Inverse probability treatment weighting resulted in a good balance of recipient covariates across values of continuous KDPI. Weighted analysis showed KDPI to be a significant predictor for short-term outcomes. A comparable (in terms of age, time on dialysis, previous transplants, gender, diabetes status, computed panel-reactive antibodies, and HLA mismatches) average recipient, receiving a kidney from a donor with KDPI 40-60 had a 3.5% risk of EGL increased to a risk of 7.5% if received a kidney from a KDPI >95 donor (hazard ratio, 2.3; 95% confidence interval, 1.9-2.7). However, for all-cause survival KDPI was less influential. The predictive ability of KDPI does not stem from recipient confounding alone. In this large sample-sized study, modeling methods accounting for nonindependence of recipient selection verify graft quality to effectively predict short-term transplantation outcomes.

Clinical Outcomes in Type 2 Diabetes Patients After Acute Myocardial Infarction: AComparison of Sodium-Glucose Cotransporter 2 Inhibitors vs. Non-Users.

To compare clinical outcomes in patients with type 2 diabetes (T2D) after acute myocardial infarction (AMI) using sodium-glucose cotransporter-2 inhibitors (SGLT2i) vs. non-use of SGLT2i. A national cohort study based on the Taiwan National Health Insurance Research Database enrolled 944 patients with T2D who had experienced AMI and were treated with SGLT2i and 8,941 patients who did not receive SGLT2i, respectively, from May 1, 2016, to December 31, 2019. We used propensity score matching to balance covariates across study groups. The follow-up period was from the index date to the independent occurrence of the study outcomes, discontinuation of the index drug, or the end of the study period (December 31, 2020), whichever occurred first. The SGLT2i group exhibited a significantly lower incidence of cardiovascular death (0.865% per year vs. 2.048% per year; hazard ratio (HR): 0.42; 95% confidence interval (CI): 0.24-0.76; P = 0.0042), heart failure hospitalization (1.987% per year vs. 3.395% per year; HR: 0.59; 95% CI: 0.39-0.89; P = 0.0126), and all-cause mortality (3.406% per year vs. 4.981% per year, HR: 0.69; 95% CI: 0.50-0.95; P = 0.0225) compared with the non-SGLT2i group. There were no significant differences between the two groups in the incidence of AMI, ischemic stroke, coronary revascularization, major adverse cardiovascular events, composite renal outcomes, or lower limb amputation. These findings suggest that the use of SGLT2i may have favorable effects on clinical outcomes in patients with T2D after AMI.

Propensity Score Analysis With Unreliable Covariates: A Comparison of Five Reliability-Adjustment Methods

Propensity score analysis (PSA) is a crucial tool for researchers in mitigating selection bias arising from multiple covariates in quasi-experimental studies. Nevertheless, the impact of low-reliability covariates on PSA necessitates careful consideration. This study employs Monte Carlo simulation to assess five methods to adjust propensity scores for unreliability of covariates. The findings reveal that the latent variable model incorporating inclusive factor scores (PSIF) results in the smallest relative bias of treatment effect estimates. Notably, only PSIF consistently provides unbiased treatment effect estimates across all conditions. Furthermore, the study underscores the potential for a misleading covariate balance evaluation when dealing with unreliable covariates, given that treatment effect estimates may be biased even when the covariate balance is perceived as adequate.

Association of nirmatrelvir–ritonavir with post-acute sequelae and mortality in patients admitted to hospital with COVID-19: a retrospective cohort study

Studies have established the short-term efficacy of nirmatrelvir-ritonavir in managing COVID-19, yet its effect on post-COVID-19 condition, especially in patients admitted to hospital, remains understudied. This study aimed to examine the effect of nirmatrelvir-ritonavir on post-COVID-19 condition among patients admitted to hospital in Hong Kong. This retrospective cohort study used real-world, territory-wide inpatient records, vaccination records, and confirmed COVID-19 case data from the Hong Kong Hospital Authority and Department of Health, The Government of the Hong Kong Special Administrative Region. Patients aged 18 years and older who tested positive for SARS-CoV-2 between March 11, 2022, and Oct 10, 2023, and who were admitted to hospital with COVID-19 were included. The treatment group included patients prescribed nirmatrelvir-ritonavir within 5 days of symptom onset, excluding those prescribed molnupiravir within 21 days, and the control group had no exposure to either nirmatrelvir-ritonavir or molnupiravir. The outcomes were post-acute inpatient death and 13 sequelae (congestive heart failure, atrial fibrillation, coronary artery disease, deep vein thrombosis, chronic pulmonary disease, acute respiratory distress syndrome, interstitial lung disease, seizure, anxiety, post-traumatic stress disorder, end-stage renal disease, acute kidney injury, and pancreatitis). These outcomes were evaluated starting at 21 days after the positive RT-PCR date in each respective cohort constructed for the outcome. Standardised mortality ratio weights were applied to balance covariates, and Cox proportional hazards regression was used to investigate the relationship between nirmatrelvir-ritonavir and outcomes. 136 973 patients were screened for inclusion, among whom 50 055 were eligible and included in the analysis (24 873 [49·7%] were female and 25 182 [50·3%] were male). 15 242 patients were prescribed nirmatrelvir-ritonavir during acute COVID-19 and 23 756 patients were included in the control group; 11 057 patients did not meet our definition for the exposed and unexposed groups. Patients were followed up for a median of 393 days (IQR 317-489). In the nirmatrelvir-ritonavir group compared with the control group, there was a significantly lower hazard of post-acute inpatient death (hazard ratio 0·62 [95% CI 0·57-0·68]; p<0·0001), congestive heart failure (0·70 [0·58-0·85]; p=0·0002), atrial fibrillation (0·63 [0·52-0·76]; p<0·0001), coronary artery disease (0·71 [0·59-0·85]; p=0·0002), chronic pulmonary disease (0·68 [0·54-0·86]; p=0·0011), acute respiratory distress syndrome (0·71 [0·58-0·86]; p=0·0007), interstitial lung disease (0·17 [0·04-0·75]; p=0·020), and end-stage renal disease (0·37 [0·18-0·74]; p=0·0049). There was no evidence indicating difference between the groups in deep vein thrombosis, seizure, anxiety, post-traumatic stress disorder, acute kidney injury, and pancreatitis. This study showed extended benefits of nirmatrelvir-ritonavir for reducing the risk of post-acute inpatient death as well as cardiovascular and respiratory complications among patients admitted to hospital with COVID-19. Further research is essential to uncover the underlying mechanisms responsible for these observed negative associations and to devise effective strategies for preventing the onset of post-acute sequelae. Health and Medical Research Fund, Research Grants Council theme-based research schemes, and Research Grants Council Collaborative Research Fund.

Abstract PO5-06-03: Impact of curative-intent treatment on survival for metastatic breast cancer limited to distant lymph nodes

Abstract BACKGROUND: Metastatic breast cancer limited to distant lymph nodes is classified as stage IV but potentially represents an anatomically distinct entity amenable to curative treatment. Its optimal management is unclear, with existing studies limited by small sample sizes and short follow-up durations. We sought to investigate the impact of curative-intent treatment on breast cancer-specific survival (BCSS) and overall survival (OS) for patients with this diagnosis. METHODS: Three cohorts of patients diagnosed with breast cancer from 2010 to 2019 were identified from the Surveillance, Epidemiology, and End Results 17 registries database. The curative cohort consisted of patients with metastatic breast cancer involving only distant lymph nodes who received a combination of chemotherapy, surgery to the primary site, and radiotherapy (curative-intent treatment). The palliative cohort included patients who underwent chemotherapy but not curative-intent treatment. The N3c cohort was comprised of patients with ipsilateral supraclavicular lymph node metastases but no distant metastases who received curative-intent treatment. Weighting based on covariate balancing propensity scores was performed to balance age, sex, race, median household income, year of diagnosis, T stage, histology, pathological grade, and ER/PR/HER2 status between the cohorts. Weight-adjusted multivariable Cox regression models with the same covariates were then used to compare the BCSS and OS of the three cohorts. RESULTS: A total of 2,089 patients with a median follow-up of 37 months were eligible for analysis, of whom 577 (27.6%) were in the curative cohort, 905 (43.3%) were in the palliative cohort, and 607 (29.1%) were in the N3c cohort. After propensity score weighting, the maximal standardized mean difference for all covariates was less than 0.1. Five-year BCSS for patients in the weight-adjusted curative, palliative, and N3c cohorts were 63.0% (95% CI 58.1-68.3), 46.7% (42.4-51.4), and 66.4% (61.4-71.8), respectively; and five-year OS were 57.9% (53.0-63.2), 41.2% (37.2-45.7), and 62.2% (57.3-67.6), respectively. Compared to the palliative cohort, the curative cohort exhibited better BCSS (HR 0.47, 95% CI 0.38-0.57; P&amp;lt; 0.001) and OS (HR 0.48, 95% CI 0.40-0.58; P&amp;lt; 0.001). In contrast, the N3c cohort had similar BCSS (HR 0.93, 95% CI 0.73-1.17; P=0.53) and OS (HR 0.96, 95% CI 0.77-1.19; P=0.68) to the curative cohort. CONCLUSIONS: Patients treated curatively for metastatic breast cancer limited to distant lymph nodes had higher BCSS and OS than those receiving palliative chemotherapy and similar BCSS and OS to patients receiving curative-intent treatment for ipsilateral supraclavicular lymph node metastases. These results support the use of aggressive multimodal therapy for select stage IV breast cancer patients with only distant nodal metastases and, if confirmed, suggest that a substantial portion of such patients may be undertreated. Citation Format: Yang Xu. Impact of curative-intent treatment on survival for metastatic breast cancer limited to distant lymph nodes [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO5-06-03.

Association between COVID-19 severity and tobacco smoking status: a retrospective cohort study using propensity score matching weights analysis

IntroductionThe COVID-19 pandemic continues to be a global threat to public health, with over 766 million confirmed cases and more than 6 million reported deaths. Patients with a smoking history...

Assessing the use of the triglyceride-glycemic index (TyG), neutrophil-lymphocyte Ratio (NLR), and platelet-lymphocyte Ratio (PLR) in distinguishing benign and malignant tumors among patients with complaints of breast mass

IntroductionBreast cancer is a prevalent global health concern characterized by uncontrolled cell growth in breast tissue. In 2020, approximately 2.3 million cases were reported worldwide, with 162,468 new cases and 87,090 fatalities documented in India in 2018. Early diagnosis is crucial for reducing mortality. Our study focused on the use of markers such as the triglyceride-glycemic index and hematological markers to distinguish between benign and malignant breast masses. MethodsA prospective cross-sectional study included female patients with breast mass complaints. The target sample size was 200. Data collection included medical history, clinical breast examination, mammography, cytological assessment via fine-needle aspiration cytology (FNAC), and blood sample collection. The analyzed parameters included neutrophil-to-lymphocyte Ratio (NLR), platelet-to-lymphocyte Ratio (PLR), and triglyceride-glycemic index (TyG). Histopathological examination confirmed the FNAC results. Statistical analysis including propensity score matching, Kolmogorov–Smirnov tests, Mann–Whitney U tests, receiver's operator curve (ROC) analysis, and logistic regression models was conducted using SPSS and R Software. Additional validation was performed on 25 participants. ResultsThis study included 200 participants. 109 had benign tumors and 91 had malignant tumors. Propensity score matching balanced covariates. NLR did not significantly differ between the groups, while PLR and TyG index differed significantly. NLR correlated strongly with the breast cancer stage, but not with the BI-RADS score. PLR and TyG index showed moderate positive correlations with the BI-RADS score. ROC analysis was used to determine the optimal cutoff values for PLR and TyG index. Logistic regression models combining PLR and TyG index significantly improved malignancy prediction. ConclusionsTyG index and PLR show potential as adjunctive markers for distinguishing breast masses. NLR correlated with cancer stage but not lesion type. Combining TyG and PLR improves prediction, aiding clinical decisions, but large-scale multicenter trials and long-term validation are required for clinical implementation.

Mahalanobis balancing: A multivariate perspective on approximate covariate balancing

AbstractIn the past decade, various exact balancing‐based weighting methods were introduced to the causal inference literature. It eliminates covariate imbalance by imposing balancing constraints in a certain optimization problem, which can nevertheless be infeasible when there is bad overlap between the covariate distributions in the treated and control groups or when the covariates are high dimensional. Recently, approximate balancing was proposed as an alternative balancing framework. It resolves the feasibility issue by using inequality moment constraints instead. However, it can be difficult to select the threshold parameters. Moreover, moment constraints may not fully capture the discrepancy of covariate distributions. In this paper, we propose Mahalanobis balancing to approximately balance covariate distributions from a multivariate perspective. We use a quadratic constraint to control overall imbalance with a single threshold parameter, which can be tuned by a simple selection procedure. We show that the dual problem of Mahalanobis balancing is an norm‐based regularized regression problem, and establish interesting connection to propensity score models. We derive asymptotic properties, discuss the high‐dimensional scenario, and make extensive numerical comparisons with existing balancing methods.

OA42 Effect of intra-articular corticosteroid injections for osteoarthritis on subsequent use of pain medications: an instrumental variable and propensity score analysis

Abstract Background/Aims Intra-articular corticosteroid injections (IACI) are recommended in the UK by NICE as an adjunctive treatment for uncontrolled pain from osteoarthritis (OA). NICE guidance states there is no evidence for therapeutic benefit lasting beyond three months, and that more evidence is needed on effectiveness at non-knee joints. We therefore aimed to assess the longer-term effect of IACI use on pain medication prescriptions among people with hand, hip, knee or shoulder OA. Methods We conducted a cohort study using UK primary care (CPRD GOLD) data for patients aged ≥20 years with a first diagnosis of OA during 2005-2019, excluding those with prior orthopaedic surgery referral or OA diagnosed at multiple joints. Exposure to IACI was defined using Read codes. The primary outcome was incident prescribing of pain medication: oral non-steroidal anti-inflammatory drugs (NSAIDs), opioids, oral corticosteroids, paracetamol, partial opioids, and topical NSAIDs. In main analyses, an instrumental variable (IV) approach was used. Index date was six months after OA diagnosis date and IV was defined as practice preference for IACI use in the year prior to index date (single and recurrent use considered separately). Outcomes were measured over five years from index date. Covariate balance was assessed, and two-step Poisson regression used where assumptions met. In secondary analyses, propensity score (PS) matching was carried out and time-varying exposure Cox models used to estimate treatment effect. Results Of 181,818 eligible patients, 92,157 (50.7%) were retained in IV analyses of single IACI and 89,103 (49.0%) of repeat IACI. Covariates were well balanced (SMD ≤0.1) across IV groups for hand (single IACI only) and knee OA, but not hip or shoulder OA. In IV analyses of hand OA, single IACI was associated with lower 5-year cumulative incidence of several pain medications including oral corticosteroids (relative risk [RR] 0.27 [95% CI 0.10, 0.76], paracetamol (RR 0.27 [0.09, 0.78] and partial opioids (RR 0.24 [0.10, 0.60]). IACI for knee OA was associated with lower 5-year cumulative incidence of most pain medications, e.g., for partial opioids a RR 0.46 [95% CI 0.34, 0.61] and RR 0.33 [0.21, 0.51] for single and recurrent IACI, respectively. Most PS matched analyses had &amp;gt;20,000 included patients (all joints combined), with covariates well balanced across IACI exposure status. PS matching confirmed a reduction in incidence of several pain medications associated with single IACI, e.g., lower incidence of partial opioids following IACI for any of hand, hip, knee or shoulder OA. Conclusion Our main findings suggest IACI use for hand or knee OA may reduce subsequent need for analgesics. Among knee OA patients, generally larger reductions were observed following recurrent than single IACI. In secondary analyses, single IACI for hand, hip, knee or shoulder OA was associated with lower subsequent use of partial opioids. Disclosure S. Hawley: None. A. Prats-Uribe: Grants/research support; European Medicines Agency. G. Matharu: Grants/research support; National Institute of Health and Care Research, Academy of Medical Sciences. A. Delmestri: None. D. Prieto-Alhambra: Consultancies; AstraZeneca, UCB Biopharma. Grants/research support; European Medicines Agency, Innovative Medicines Initiative, Amgen, Chiesi-Taylor, Lilly, Janssen, Novartis, UCB Biopharma. Other; Amgen, Astrellas, Janssen, Synapse Management Partners, UCB Biopharma. A. Judge: Grants/research support; National Institute of Health and Care Research, Health Data Research UK, Versus Arthritis, Healthcare Quality Improvement Partnership (HQIP), Royal College of Physicians (RCP), Tommy's, The Health Foundation. M. Whitehouse: Royalties; Taylor Francis. Member of speakers’ bureau; Heraeus. Grants/research support; National Institute of Health and Care Research, Healthcare Quality Improvement Partnership (HQIP), Ceramtec.

Energy balancing of covariate distributions

AbstractBias in causal comparisons has a correspondence with distributional imbalance of covariates between treatment groups. Weighting strategies such as inverse propensity score weighting attempt to mitigate bias by either modeling the treatment assignment mechanism or balancing specified covariate moments. This article introduces a new weighting method, called energy balancing, which instead aims to balance weighted covariate distributions. By directly targeting distributional imbalance, the proposed weighting strategy can be flexibly utilized in a wide variety of causal analyses without the need for careful model or moment specification. Our energy balancing weights (EBW) approach has several advantages over existing weighting techniques. First, it offers a model-free and robust approach for obtaining covariate balance that does not require tuning parameters, obviating the need for modeling decisions of secondary nature to the scientific question at hand. Second, since this approach is based on a genuine measure of distributional balance, it provides a means for assessing the balance induced by a given set of weights for a given dataset. We demonstrate the effectiveness of this EBW approach in a suite of simulation experiments, and in studies on the safety of right heart catheterization and on three additional studies using electronic health record data.

Assessing the effect of remdesivir alone and in combination with corticosteroids on time to death in COVID-19: A propensity score-matched analysis

BackgroundThis study aimed to evaluate the impact of remdesivir, alone or in combination with corticosteroids, on the time to death in COVID-19 patients. MethodsThis retrospective cohort study was conducted between March 20, 2020, and March 18, 2021, on 5318 patients with confirmed COVID-19 at three tertiary educational hospitals in Tehran, Iran. The treated group consisted of COVID-19 patients who received remdesivir, either alone or in combination with corticosteroids, while the untreated group included infected patients who did not receive these treatments. The two groups were matched by propensity score matching method, which adjusted for 92 confounding factors such as demographic, sign and symptom, clinical, and treatment-related factors, to provide balanced covariates between the treated and untreated groups. ResultsOf the 4,322 patients, 791 received remdesivir alone and 694 received remdesivir combined with corticosteroids. Among the patients who received remdesivir, 456 were male, and the mean (±standard deviation) age was 59.52±16.49 with a median hospitalization stay of 8 (interquartile range: 5–13) days. After applying the nearest neighbor propensity score matching method, there were no significant differences found in the hazard of death between patients who received remdesivir alone and those who did not receive treatment (hazard ratio: 0.94; 95 % confidence interval: 0.76, 1.16). In addition, the combination of remdesivir and corticosteroids was not associated with a reduced hazard of death (hazard ratio: 1.00; 95 % confidence interval: 0.80, 1.26). The median hospitalization stay was significantly longer in the group treated with remdesivir (alone/combined with corticosteroids) compared to the untreated group, both before and after matching. ConclusionNo association was observed between remdesivir treatment (alone/combined with corticosteroids) and a reduced hazard of death among COVID-19 patients. However, longer hospitalization stays were observed in patients receiving remdesivir, either alone or in combination with corticosteroids, compared to the untreated group.

Estimating dynamic treatment regimes for ordinal outcomes with household interference: Application in household smoking cessation.

The focus of precision medicine is on decision support, often in the form of dynamic treatment regimes, which are sequences of decision rules. At each decision point, the decision rules determine the next treatment according to the patient's baseline characteristics, the information on treatments and responses accrued by that point, and the patient's current health status, including symptom severity and other measures. However, dynamic treatment regime estimation with ordinal outcomes is rarely studied, and rarer still in the context of interference - where one patient's treatment may affect another's outcome. In this paper, we introduce the weighted proportional odds model: a regression based, approximate doubly-robust approach to single-stage dynamic treatment regime estimation for ordinal outcomes. This method also accounts for the possibility of interference between individuals sharing a household through the use of covariate balancing weights derived from joint propensity scores. Examining different types of balancing weights, we verify the approximate double robustness of weighted proportional odds model with our adjusted weights via simulation studies. We further extend weighted proportional odds model to multi-stage dynamic treatment regime estimation with household interference, namely dynamic weighted proportional odds model. Lastly, we demonstrate our proposed methodology in the analysis of longitudinal survey data from the Population Assessment of Tobacco and Health study, which motivates this work. Furthermore, considering interference, we provide optimal treatment strategies for households to achieve smoking cessation of the pair in the household.

On variance estimation of the inverse probability-of-treatment weighting estimator: A tutorial for different types of propensity score weights.

Propensity score methods, such as inverse probability-of-treatment weighting (IPTW), have been increasingly used for covariate balancing in both observational studies and randomized trials, allowing the control of both systematic and chance imbalances. Approaches using IPTW are based on two steps: (i) estimation of the individual propensity scores (PS), and (ii) estimation of the treatment effect by applying PS weights. Thus, a variance estimator that accounts for both steps is crucial for correct inference. Using a variance estimator which ignores the first step leads to overestimated variance when the estimand is the average treatment effect (ATE), and to under or overestimated estimates when targeting the average treatment effect on the treated (ATT). In this article, we emphasize the importance of using an IPTW variance estimator that correctly considers the uncertainty in PS estimation. We present a comprehensive tutorial to obtain unbiased variance estimates, by proposing and applying a unifying formula for different types of PS weights (ATE, ATT, matching and overlap weights). This can be derived either via the linearization approach or M-estimation. Extensive R code is provided along with the corresponding large-sample theory. We perform simulation studies to illustrate the behavior of the estimators under different treatment and outcome prevalences and demonstrate appropriate behavior of the analytical variance estimator. We also use a reproducible analysis of observational lung cancer data as an illustrative example, estimating the effect of receiving a PET-CT scan on the receipt of surgery.

Evaluating the comparability of osteoporosis treatments using propensity score and negative control outcome methods in UK and Denmark electronic health record databases.

Evidence on the comparative effectiveness of osteoporosis treatments is heterogeneous. This may be attributed to different populations and clinical practice, but also to differing methodologies ensuring comparability of treatment groups before treatment effect estimation and the amount of residual confounding by indication. This study assessed the comparability of denosumab vs oral bisphosphonate (OBP) groups using propensity score (PS) methods and negative control outcome (NCO) analysis. A total of 280 288 women aged ≥50yr initiating denosumab or OBP in 2011-2018 were included from the UK Clinical Practice Research Datalink (CPRD) and the Danish National Registries (DNR). Balance of observed covariates was assessed using absolute standardized mean difference (ASMD) before and after PS weighting, matching, and stratification, with ASMD >0.1 indicating imbalance. Residual confounding was assessed using NCOs with ≥100 events. Hazard ratio (HR) and 95%CI between treatment and NCO were estimated using Cox models. Presence of residual confounding was evaluated with 2 approaches (1) >5% of NCOs with 95% CI excluding 1, (2) >5% of NCOs with an upper CI <0.75 or lower CI >1.3. The number of imbalanced covariates before adjustment (CPRD 22/87; DNR 18/83) decreased, with 2%-11% imbalance remaining after weighting, matching, or stratification. Using approach 1, residual confounding was present for all PS methods in both databases (≥8% of NCOs), except for stratification in DNR (3.8%). Using approach 2, residual confounding was present in CPRD with PS matching (5.3%) and stratification (6.4%), but not with weighting (4.3%). Within DNR, no NCOs had HR estimates with upper or lower CI limits beyond the specified bounds indicating residual confounding for any PS method. Achievement of covariate balance and determination of residual bias were dependent upon several factors including the population under study, PS method, prevalence of NCO, and the threshold indicating residual confounding.

Covariate Balancing Methods for Randomized Controlled Trials Are Not Adversarially Robust.

The first step toward investigating the effectiveness of a treatment via a randomized trial is to split the population into control and treatment groups then compare the average response of the treatment group receiving the treatment to the control group receiving the placebo. To ensure that the difference between the two groups is caused only by the treatment, it is crucial that the control and the treatment groups have similar statistics. Indeed, the validity and reliability of a trial are determined by the similarity of two groups' statistics. Covariate balancing methods increase the similarity between the distributions of the two groups' covariates. However, often in practice, there are not enough samples to accurately estimate the groups' covariate distributions. In this article, we empirically show that covariate balancing with the standardized means difference (SMD) covariate balancing measure, as well as Pocock and Simon's sequential treatment assignment method, are susceptible to worst case treatment assignments. Worst case treatment assignments are those admitted by the covariate balance measure, but result in highest possible ATE estimation errors. We developed an adversarial attack to find adversarial treatment assignment for any given trial. Then, we provide an index to measure how close the given trial is to the worst case. To this end, we provide an optimization-based algorithm, namely adversarial treatment assignment in treatment effect trials (ATASTREET), to find the adversarial treatment assignments.