Comparing Bayesian hierarchical meta-regression methods and evaluating the influence of priors for evaluations of surrogate endpoints on heterogeneous collections of clinical trials

BACKGROUND

Surrogate endpoints, such as those of interest in chronic kidney disease (CKD), are often evaluated using Bayesian meta-regression. Trials used for the analysis can evaluate a variety of interventions for different sub-classifications of disease, which can introduce two additional goals in the analysis. The first is to infer the quality of the surrogate within specific trial subgroups defined by disease or intervention classes. The second is to generate more targeted subgroup-specific predictions of treatment effects on the clinical endpoint.

METHODS

Using real data from a collection of CKD trials and a simulation study, we contrasted surrogate endpoint evaluations under different hierarchical Bayesian approaches. Each approach we considered induces different assumptions regarding the relatedness (exchangeability) of trials within and between subgroups. These include partial-pooling approaches, which allow subgroup-specific meta-regressions and, yet, facilitate data adaptive information sharing across subgroups to potentially improve inferential precision. Because partial-pooling models come with additional parameters relative to a standard approach assuming one meta-regression for the entire set of studies, we performed analyses to understand the impact of the parameterization and priors with the overall goals of comparing precision in estimates of subgroup-specific meta-regression parameters and predictive performance.

RESULTS

In the analyses considered, partial-pooling approaches to surrogate endpoint evaluation improved accuracy of estimation of subgroup-specific meta-regression parameters relative to fitting separate models within subgroups. A random rather than fixed effects approach led to reduced bias in estimation of meta-regression parameters and in prediction in subgroups where the surrogate was strong. Finally, we found that subgroup-specific meta-regression posteriors were robust to use of constrained priors under the partial-pooling approach, and that use of constrained priors could facilitate more precise prediction for clinical effects in trials of a subgroup not available for the initial surrogacy evaluation.

CONCLUSION

Partial-pooling modeling strategies should be considered for surrogate endpoint evaluation on collections of heterogeneous studies. Fitting these models comes with additional complexity related to choosing priors. Constrained priors should be considered when using partial-pooling models when the goal is to predict the treatment effect on the clinical endpoint.

Effect of carvedilol versus placebo on cardiac function in anthracycline-exposed survivors of childhood cancer (PREVENT-HF): a randomised, controlled, phase 2b trial

BACKGROUND

Carvedilol improves cardiac function in patients with heart failure but remains untested as cardioprotective therapy in long-term childhood cancer survivors (ie, those who have completed treatment for childhood cancer and are in remission) at risk for heart failure due to high-dose anthracycline exposure. We aimed to evaluate the activity and safety of low-dose carvedilol for heart failure risk reduction in childhood cancer survivors at highest risk for heart failure.

METHODS

PREVENT-HF was a randomised, double-blind, phase 2b trial done at 30 hospitals in the USA and Canada. Patients were eligible if they had any cancer diagnosis that resulted in at least 250 mg/m2 cumulative exposure to anthracycline by age 21 years; completed their cancer treatment at least 2 years previously; an ejection fraction of at least 50% or fractional shortening of at least 25%, or both; and bodyweight of at least 40 kg. Patients were randomly assigned (1:1) with automated computer-generated permuted block randomisation (block size of 4), stratified by age at diagnosis, time since diagnosis, and history of chest-directed radiotherapy, to carvedilol (up-titrated from 3·125 g per day to 12·5 mg per day) or placebo orally for 2 years. Participants, staff, and investigators were masked to study group allocation. The primary endpoint was to establish the effect of carvedilol on standardised left ventricular wall thickness-dimension ratio Z score (LVWT/Dz). Treatment effects were analysed with a linear mixed-effects model for normally distributed data with a linear time effect and testing the significance of treatment*time interaction in the modified intention-to-treat (mITT) cohort (ie, all randomly assigned participants who had a baseline and at least one subsequent echocardiogram measurement). Safety was assessed in the ITT population (ie, all randomly assigned participants). This trial was registered with ClinicalTrials.gov, NCT027175073, and enrolment and follow-up are complete.

FINDINGS

Between July 3, 2012, and June 22, 2020, 196 participants were enrolled, of whom 182 (93%) were eligible and randomly assigned to either carvedilol (n=89) or placebo (n=93; ITT population). Median age was 24·7 years (IQR 19·6-36·6), 91 (50%) participants were female, 91 (50%) were male, and 119 (65%) were non-Hispanic White. As of data cutoff (June 10, 2022), median follow-up was 725 days (IQR 378-730). 151 (n=75 in the carvedilol group and n=76 in the placebo group) of 182 participants were included in the mITT population, among whom LVWT/Dz was similar between the two groups (-0·14 [95% CI -0·43 to 0·16] in the carvedilol group vs -0·45 [-0·77 to -0·13] in the placebo group; difference 0·31 [95% CI -0·10 to 0·73]; p=0·14). Two (2%) of 89 patients in the carvedilol group two adverse events of grade 2 or higher (n=1 shortness of breath and n=1 arthralgia) and none in the placebo group. There were no adverse events of grade 3 or higher and no deaths.

INTERPRETATION

Low-dose carvedilol appears to be safe in long-term childhood cancer survivors at risk for heart failure, but did not result in significant improvement of LVWT/Dz compared with placebo. These results do not support the use of carvedilol for secondary heart failure prevention in anthracycline-exposed childhood cancer survivors.

FUNDING

National Cancer Institute, Leukemia & Lymphoma Society, St Baldrick's Foundation, Altschul Foundation, Rally Foundation, American Lebanese Syrian Associated Charities.

Handling missing within-study correlations in the evaluation of surrogate endpoints

Rigorous evaluation of surrogate endpoints is performed in a trial-level analysis in which the strength of the association between treatment effects on the clinical and surrogate endpoints is quantified across a collection of previously conducted trials. To reduce bias in measures of the performance of the surrogate, the statistical model must account for the sampling error in each trial's estimated treatment effects and their potential correlation. Unfortunately, these within-study correlations can be difficult to obtain, especially for meta-analysis of published trial results where individual patient data is not available. As such, these terms are frequently partially or completely missing in the analysis. We show that improper handling of these missing terms can meaningfully alter the perceived quality of the surrogate and we introduce novel strategies to handle the missingness.

A meta-analysis of GFR slope as a surrogate endpoint for kidney failure

Glomerular filtration rate (GFR) decline is causally associated with kidney failure and is a candidate surrogate endpoint for clinical trials of chronic kidney disease (CKD) progression. Analyses across a diverse spectrum of interventions and populations is required for acceptance of GFR decline as an endpoint. In an analysis of individual participant data, for each of 66 studies (total of 186,312 participants), we estimated treatment effects on the total GFR slope, computed from baseline to 3 years, and chronic slope, starting at 3 months after randomization, and on the clinical endpoint (doubling of serum creatinine, GFR < 15 ml min-1 per 1.73 m2 or kidney failure with replacement therapy). We used a Bayesian mixed-effects meta-regression model to relate treatment effects on GFR slope with those on the clinical endpoint across all studies and by disease groups (diabetes, glomerular diseases, CKD or cardiovascular diseases). Treatment effects on the clinical endpoint were strongly associated with treatment effects on total slope (median coefficient of determination (R2) = 0.97 (95% Bayesian credible interval (BCI) 0.82-1.00)) and moderately associated with those on chronic slope (R2 = 0.55 (95% BCI 0.25-0.77)). There was no evidence of heterogeneity across disease. Our results support the use of total slope as a primary endpoint for clinical trials of CKD progression.

Evaluation of Variation in the Performance of GFR Slope as a Surrogate End Point for Kidney Failure in Clinical Trials that Differ by Severity of CKD

BACKGROUND

The GFR slope has been evaluated as a surrogate end point for kidney failure in meta-analyses on a broad collection of randomized controlled trials (RCTs) in CKD. These analyses evaluate how accurately a treatment effect on GFR slope predicts a treatment effect on kidney failure. We sought to determine whether severity of CKD in the patient population modifies the performance of GFR slope.

METHODS

We performed Bayesian meta-regression analyses on 66 CKD RCTs to evaluate associations between effects on GFR slope (the chronic slope and the total slope over 3 years, expressed as mean differences in ml/min per 1.73 m2/yr) and those of the clinical end point (doubling of serum creatinine, GFR <15 ml/min per 1.73 m2, or kidney failure, expressed as a log-hazard ratio), where models allow interaction with variables defining disease severity. We evaluated three measures (baseline GFR in 10 ml/min per 1.73 m2, baseline urine albumin-to-creatinine ratio [UACR] per doubling in mg/g, and CKD progression rate defined as the control arm chronic slope, in ml/min per 1.73 m2/yr) and defined strong evidence for modification when 95% posterior credible intervals for interaction terms excluded zero.

RESULTS

There was no evidence for modification by disease severity when evaluating 3-year total slope (95% credible intervals for the interaction slope: baseline GFR [-0.05 to 0.03]; baseline UACR [-0.02 to 0.04]; CKD progression rate [-0.07 to 0.02]). There was strong evidence for modification in evaluations of chronic slope (95% credible intervals: baseline GFR [0.02 to 0.11]; baseline UACR [-0.11 to -0.02]; CKD progression rate [0.01 to 0.15]).

CONCLUSIONS

These analyses indicate consistency of the performance of total slope over 3 years, which provides further evidence for its validity as a surrogate end point in RCTs representing varied CKD populations.

Characteristics of Nonelderly Adult Health Care Persistent Super Utilizers in Utah

In the United States, the top 1% and top 5% of health care spenders account for 23% and 50% of total health care spending, respectively. These high spenders have been coined the term super utilizers (SU). The aim of this study was to identify the characteristics associated with these patients to aid in developing public health interventions aimed at transitioning patients out of the SU category and thus ultimately helping to control health care costs. The authors utilized the Utah All-Payer Claims Database and Utah Population Database from 2013 to 2015 to identify demographics, comorbid conditions, health care utilization, and cost characteristics of persistent super utilizers (PSU) (≥3 hospitalizations per year for 3 years) of health care compared with persistent nonsuper utilizers (PNSU) (<3 hospitalizations per year for 3 years). Multivariable logistic regression was utilized to identify the characteristics associated with PSU versus PNSU. Higher outpatient/Emergency Department/noninpatient (eg, visits with imaging and Centers for Medicare & Medicaid Services preventive visits) health care utilization and spending, and prevalence of comorbid disease and psychosocial conditions were associated with PSU. In multivariable analysis, factors such as heart disease, chronic kidney disease (CKD), diabetes, alcohol abuse, and depression were statistically significantly associated with higher odds of PSU, with the most noteworthy being CKD (odds ratio [OR] 6.85, 95% confidence interval [95% CI] 5.84-8.02; P < 0.001), alcohol abuse (OR 5.90, 95% CI 4.49-7.69; P < 0.001), and heart diseases (OR 4.41, 95% CI 3.74-5.18; P < 0.001). The annual health care cost of a PSU is about 11.5 times greater than a PNSU ($54,776 vs. $4801; P < 0.001).

Patient-reported opioid use for tissue expander-based breast reconstruction

INTRODUCTION

There is limited evidence for appropriate post-operative opioid prescribing in breast reconstruction patients. We sought to describe postoperative outpatient prescription opioid use patterns (quantity and duration) following discharge after immediate breast reconstruction with tissue expanders (TE) and to identify demographic and/or clinical risk factors associated with postoperative outpatient opioid use.

METHODS

Patients 18 years and older undergoing immediate TE-based breast reconstruction were given a 28-day postoperative pain medication log book. Descriptive statistics were performed to describe the quantity and duration of opioid use. Preoperative, intraoperative, and postoperative characteristics were examined and tested for their associations with postoperative opioid use.

RESULTS

A total of 45 logbooks were completed. On average, patients used opioids for 7.42 days (SD = 6.45) after discharge home and used 15.9 (SD = 18.71) oxycodone 5 mg tablet equivalents (119.3 morphine milligram equivalents, SD = 140.31). The total number of oxycodone 5 mg equivalents consumed prior to discharge was associated with the amount of post-discharge opioid consumption (IRR=1.08, p<0.01). Each additional year of age was associated with a reduction in the days-to-opioid cessation by a factor of 0.97 (p=0.01). Each additional oxycodone 5mg equivalent consumed prior to hospital discharge was associated with an increase in the days-to-cessation after discharge by a factor of 1.04 (p=0.026).

CONCLUSIONS

These patient-reported data will provide a benchmark which plastic surgeons can use to minimize narcotic use in patients and will help prevent issues of dependence, misuse, and diversion, while being mindful of adequate pain control. For patients discharging home after a one-night stay for immediate TE breast reconstruction, we recommend a prescription for 10 oxycodone 5 mg tablets, or 15 tablets if they are less than age 49 or have had high inpatient opioid use. Patients should also be counseled that the expected duration of outpatient opioid use is 7-11 days, and that 20 % of patients did not use any opioids following hospital discharge, making nonnarcotic pain regimens a real possibility.

Identifying predictors of ventral hernia recurrence: systematic review and meta-analysis

BACKGROUND

Ventra hernias are increasing in prevalence and many recur despite attempted repair. To date, much of the literature is underpowered and divergent. As a result there is limited high quality evidence to inform surgeons succinctly which perioperative variables influence postoperative recurrence. This systematic review aimed to identify predictors of ventral hernia recurrence.

METHODS

PubMed was searched for studies reporting prognostic data of ventral hernia recurrence between 1 January 1995 and 1 January 2018. Extracted data described hernia type (primary/incisional), definitions of recurrence, methods used to detect recurrence, duration of follow-up, and co-morbidity. Data were extracted for all potential predictors, estimates and thresholds described. Random-effects meta-analysis was used. Bias was assessed with a modified PROBAST (Prediction model Risk Of Bias ASsessment Tool).

RESULTS

Screening of 18 214 abstracts yielded 274 individual studies for inclusion. Hernia recurrence was defined in 66 studies (24.1 per cent), using 41 different unstandardized definitions. Three patient variables (female sex, age 65 years or less, and BMI greater than 25, 30, 35 or 40 kg/m2), five patient co-morbidities (smoking, diabetes, chronic obstructive pulmonary disease, ASA grade III-IV, steroid use), two hernia-related variables (incisional/primary, recurrent/primary), six intraoperative variables (biological mesh, bridged repair, open versus laparoscopic surgery, suture versus mesh repair, onlay/retrorectus, intraperitoneal/retrorectus), and six postoperative variables (any complication, surgical-site occurrence, wound infection, seroma, haematoma, wound dehiscence) were identified as significant prognostic factors for hernia recurrence.

CONCLUSION

This study summarized the current evidence base for predicting ventral hernia recurrence. Results should inform best practice and future research.

A Nationwide Analysis of Early and Late Readmissions following Free Tissue Transfer for Breast Reconstruction

BACKGROUND

 Traditionally, surgical quality outcomes are assessed using a 30-day postoperative window. For breast cancer patients undergoing free tissue transfer for breast reconstruction, we sought to describe the distribution of and specific risk factors for early and late readmissions within a 0- to 90-day postoperative period.

PATIENTS AND METHODS

 The Nationwide Readmissions Database was used to conduct a retrospective cohort study. Breast cancer patients undergoing free tissue transfer for breast reconstruction were identified using International Classification of Diseases -9 diagnosis and procedure codes. Ninety-day readmissions related to infection or wound complications were identified. Univariable and multivariable logistic regression models were used to identify patient risk factors for readmissions that occurred early (0-30 days) and late (31-90 days) after their index procedure.

RESULTS

 In the weighted sample, we identified approximately 7,305 free flap breast reconstructions and a surgical wound-related readmission rate of 4.3% (n = 312): 65.4% of the readmissions occurred early while 34.6% occurred late after surgery. The mean days to readmission was 26, and 75% of all readmissions occurred within the first 36 days after surgery. Variables independently associated with readmissions during the 0- to 90-day postoperative period included: history of chronic obstructive pulmonary disease (p = 0.036), hypertension (p = 0.03), obesity (p ≤ 0.001), and history of smoking (p = 0.004). The variables independently associated with the early readmission period were the same as those identified for the 0- to 90-day postoperative period. The variables independently associated with late readmissions were different: history of depression (p = 0.001) and history of smoking (p = 0.001).

CONCLUSION

 The conventional 30-day hospital readmission rate classically used as a quality metric is overlooking a significant portion of admissions after free flap-based breast reconstruction. Different variables were found to be associated with readmission in the early versus late cohorts. Interventions targeting these variables could decrease readmissions and their associated costs.

Effect of denervation on lamellar cells of Meissner-like sensory corpuscles of the rat. An immunohistochemical study

The denervation-induced changes on S-100 protein, glial fibrillary acidic protein (GFAP) and vimentin immunoreactivity (IR) of the lamellar cells from cutaneous Meissner-like sensory nerve formations (SNF), or corpuscles, of the adult rat hind limb foot-pads were studied, using combined immunohistochemical and image analysis (optic microdensitometry) techniques. Animals were allowed to survive for 1, 3 and 7 days following sciatic and saphenous nerves transection. Lamellar cells of Meissner-like corpuscles displayed S-100 protein- and vimentin-IR, but not GFAP-IR. Denervation caused a marked time-dependent decrease of S-100 protein IR whereas vimentin-IR did not change or weakly increased. No positive GFAP-IR was observed in denervated SNF. These findings suggest that continuity of SNF with nerve fibers supplying them is necessary to maintain some of the immunohistochemical characteristics of the non-neuronal cells of SNF.

Prophylactic tranexamic acid decreases bleeding after cardiac operations

Thirty-eight patients undergoing a cardiac operation randomly received either tranexamic acid, a potent inhibitor of plasminogen, or placebo in an effort to determine whether prophylactic antifibrinolytic therapy reduces chest tube drainage. Twelve-hour blood loss was 750 +/- 314 (standard deviation) ml in the placebo group and 496 +/- 228 ml in the drug group (p = 0.0057). Fibrin split products were present more frequently in patients in the placebo group (17 of 20 compared with four of 18 in the drug group; p = 0.0002). Tranexamic acid markedly decreased plasminogen availability (112 +/- 104 units in the placebo group versus 36 +/- 18 units in the drug group, p = 0.0058). Plasma fibrinogen concentrations were similar in the placebo and drug groups. Patients in the placebo group received more fresh-frozen plasma and more mediastinal shed blood than those in the drug group. No coagulation-related complication occurred in the group receiving tranexamic acid. We conclude that prophylactic tranexamic acid can be administered safely to inhibit fibrinolysis during cardiac operations, decrease postoperative bleeding, and possibly decrease the frequency of blood product transfusion.

Kinetics of Acid-Hydrolysis of Nickel(II) and Copper(II) Compounds With the Cyclic Diamines 1,5-Diazocane and 4,4-Dimethyl-7-(5,5,7-trimethyl-1,4-diazepan-1-Yl)-5-azaheptan-2-ol

The kinetics of acid-promoted hydrolysis reactions of nickel(II) and copper(ll) complexes of the cyclic diamines 1,5-diazocane ( daco ) and 4,4-dimethyl-7-(5,5,7-trimethyl-1,4-diazepan-1-yl)-5-azaheptan-2-ol (pyaz) in NaCl/HCl media ([Cl-] = 2 mol dm-3) are reported. For [Ni( daco )2]2+, beyond small [H+], an acid-limited rate constant of 2.0×10-5 S-l, involving dissociation of the first daco ligand, and for [Cu( daco )2]2+ a similarly acid-limited rate constant of 1.5×10-2 s-l, involving dissociation of the second daco ligand, were measured at 50°C. For [Ni( pyaz )]2+ and [Cu ( pyaz )2+ the rates of the hydrolysis reactions are again acid-limited, effectively independent of acid with rate constant of 1.0×10-2 s-1 for [Ni( pyaz )]2+ while for [Cu( pyaz )]2+ the rate follows the expression kobs , = 6.0×10-2[H+]/(l+4.6[H+]) S-l, both measured at 25°C.