Population-averaged Approach Research Articles

Human-in-the-loop methods for occupant-centric building design and operation

A comfortable indoor environment should be one of the main services buildings provide. However, technical building systems are typically designed and operated according to fixed set-point temperatures determined by the ‘one-size-fits-all’ principle assuming universal thermal comfort requirements, which has been questioned in the last fifty years. Designing and implementing comfortable set-point modulations that consider occupant feedback would be beneficial in terms of increasing comfort, potentially reduce energy consumption and significantly support the clean energy transition. An exploratory study aimed at predicting the thermal preferences of human subjects exposed to a dynamic thermal environment is presented. Using data acquired from a laboratory experiment where subjects were exposed to precisely controlled thermal ramps in an ‘office-like’ climatic chamber, cluster-specific and population-averaged methods are designed to handle the group-level residual during the prediction of the thermal preference votes. The results show that both approaches are valid strategies for modelling thermal preference votes and are effective in supporting a concrete occupant-centric building design and the building’s operation. Furthermore, the population-averaged approach is suitable for the occupant-centric building design phase, where the target is an ‘average’ occupant. The cluster-specific method is best suited to meet the needs of a specific occupant and is suitable for implementation in the operational phase of the building.

Older people's experiences of dignity and support with eating during hospital stays: analytical framework, policies and outcomes

AbstractThere is growing recognition of the importance of dignity and support with eating as markers of high-quality and older-person-centred hospital services. We use data on these markers from the national Adult Inpatient Survey for England to build up statistical evidence on older people's experiences. We find that poor and inconsistent experiences of being treated with dignity and respect, and of receiving support with eating, affect a substantial proportion of inpatients across the vast majority of acute hospital trusts. There has been remarkably little change over time, although small improvements provide some grounds for optimism relating to policy developments in the period following the Francis Inquiry. Amongst people over 65, the prevalence of inconsistent and poor experiences of dignity and support with eating was higher amongst the ‘oldest of the old’ (inpatients aged over 80), individuals who experience a long-standing limiting illness or disability, and women. The highest rates of prevalence were observed amongst disabled women over 80. Perceptions of inadequate nursing quantity and quality, and lack of choice of food, stand out from logistic regression analysis as having consistent, large associations with lack of support with eating. These factors provide potential policy levers since they are within the control of hospitals to a certain extent. In drawing lessons from our analysis for inspection, regulation and monitoring, we highlight the importance of inequalities analysis – including systematic disaggregation and separate identification of at risk sub-groups (e.g. older disabled women) – rather than relying on a ‘population average approach’.

Is Post-discharge Rehabilitation Timing Associated with 90-Day Readmission in Primary Total Joint Arthroplasty?

Physical therapy (PT) is an accepted standard of care after total joint arthroplasty (TJA) and essential to maximizing joint functionality and minimizing complications that lead to readmission. However, evidence-based guidelines about appropriate post-discharge rehabilitative care are not well-defined in the orthopedic literature. We sought to determine the average timing for receiving PT rehabilitation and to evaluate the association between PT rehabilitation timing and unplanned readmission within 90days of a TJA patient being discharged home from acute care. This retrospective study examined 11,545 joint procedures using claims data for the years 2008 to 2013. Outcomes were assessed using a population-averaged approach to regression models. The average time for initiating PT was 4days for knee arthroplasty and 6days for hip arthroplasty in patients discharged home from acute care. Most patients (89%) began PT consultation or supervised exercises during the first week after discharge. The type of joint surgery considerably modified the effect of rehabilitation timing on the likelihood of readmission. Later initiation of rehabilitation was associated with a higher probability of 90-day readmission in both knee and hip arthroplasty, with the effect of rehabilitation timing being more pronounced in hip rather than knee arthroplasty 2weeks post-discharge from acute care. Timing for initiating PT may be an important modifiable factor that can affect readmission in patients discharged home from acute care after TJA. Further exploration of the role of PT timing along with other factors such as dosage and frequency among such patients is needed.

Two-part models for cost with zeros to decompose effects of covariates on probability of cost, mean nonzero cost, and mean total cost.

Health care cost data often contain many zero values, for patients who did not use any care. Two-part models with logistic models for part I, probability of use (ie, nonzero cost) and log-link models for part II, mean cost of use (ie, nonzero cost) are often used. Effects of exposures or covariates on total (marginal) cost are often of interest, and recent work has proposed useful methods. Factors that affect total cost do so through a combination of effects on probability of use and cost of use. Such a decomposition is needed to understand and act on factors that affect total cost, but little work has been done on this question. This paper presents two new methods for decomposing effects on total cost, namely, an adjusted approach based on log-binomial models for part I and a population average approach based on counterfactual arguments. Extensive simulations illustrate performance, interpretation, and robustness over a wide range of data features. The method is applied to risk-adjusted 30-day outpatient cardiac cost for patients following percutaneous coronary intervention from the Department of Veterans Affairs Clinical Assessment, Reporting, and Tracking Program. Results illustrate the simple decomposition of effects on total cost into components for probability of use and cost of use.

Tolerance Induction Program Effect Explains Variation in Wheal Size, sIgE, and IgG4 in Peanut Allergic Children

Background: Tolerance induction oral immunotherapy, unique to our centre and termed the Tolerance Induction Program (TIP), has shown promise as a safe form of treatment in peanut allergic children. Evaluations of emerging treatments for peanut allergy typically report from a population average perspective. Our study sought to compare rate of decline in peanut-skin prick testing wheal size and sIgE following one year of high dose, weekly peanut immunotherapy using the standard population average approach to findings that account for subject variation in intercepts and slopes. Methods: This is a descriptive study in 51 peanut allergic children who underwent TIP at the Translational Pulmonary and Immunology Research Centre in Long Beach, California. Post intervention reductions in peanutwheal size and sIgE were assessed using Wilcoxen signed rank test and mixed effects modelling procedures. Results: The population average approach estimated reduction in wheal size to 29% of baseline value (p<0.001), compared to 41% after adjustment for random intercepts and slopes, p<0.001. Reduction in sIgE to 30% of baseline value using the population average approach (p<0.001), compared to 46% in the random intercepts model (p<0.001), and 44% in the random intercepts and slopes model (p=0.064). Conclusions: Tolerance induction oral immunotherapy significantly reduced peanut-SPT wheal size and sIgE in peanut allergic children. The subject specific approach produced more conservative effect estimates than observed using the standard population average approach.

Automatic individual arterial input functions calculated from PCA outperform manual and population-averaged approaches for the pharmacokinetic modeling of DCE-MR images

To introduce a segmentation method to calculate an automatic arterial input function (AIF) based on principal component analysis (PCA) of dynamic contrast enhanced MR (DCE-MR) imaging and compare it with individual manually selected and population-averaged AIFs using calculated pharmacokinetic parameters. The study included 65 individuals with prostate examinations (27 tumors and 38 controls). Manual AIFs were individually extracted and also averaged to obtain a population AIF. Automatic AIFs were individually obtained by applying PCA to volumetric DCE-MR imaging data and finding the highest correlation of the PCs with a reference AIF. Variability was assessed using coefficients of variation and repeated measures tests. The different AIFs were used as inputs to the pharmacokinetic model and correlation coefficients, Bland-Altman plots and analysis of variance tests were obtained to compare the results. Automatic PCA-based AIFs were successfully extracted in all cases. The manual and PCA-based AIFs showed good correlation (r between pharmacokinetic parameters ranging from 0.74 to 0.95), with differences below the manual individual variability (RMSCV up to 27.3%). The population-averaged AIF showed larger differences (r from 0.30 to 0.61). The automatic PCA-based approach minimizes the variability associated to obtaining individual volume-based AIFs in DCE-MR studies of the prostate.

A longitudinal field trial assesing the impact of feeding waste milk containing antibiotic residues on the prevalence of ESBL-producing Escherichia coli in calves

A longitudinal field trial was carried out on a farm known to harbour cefotaximase (CTX-M)-positive Escherichia coli, in order to assess the impact of feeding waste milk containing antibiotic residues (WM+AR) on the prevalence of these bacteria in the faeces of calves. Fifty calves were alternately assigned to one of two groups at birth and fed either milk replacer (control group) or WM+AR (treatment group). Faecal samples were collected from all calves daily for the first week after enrolment, twice weekly until weaning, then weekly for a further six weeks. Environmental samples from the calf housing were collected weekly. WM+AR and powdered milk samples were examined for antibiotic residues and CTX-M-positive E. coli. Total E. coli and CTX-M-positive E. coli in faecal samples were enumerated using selective media. Regression analyses were performed on the bacterial count data using a population-averaged approach based on generalised estimating equations (GEE) to account for repeated measurements on individual calves over time.Cefquinome, a fourth generation cephalosporin, was detected in 87% of WM+AR samples at a mean concentration of 0.746mg/l. All environmental sampling locations yielded CTX-M-positive E. coli. Significantly more pen floor samples were positive in the treatment group. Calves in the treatment group shed greater numbers of CTX-M-positive E. coli than calves in the control group throughout the study, and shedding decreased at a slower rate in the treatment group. CTX-M-positive E. coli persisted in a larger number of calves fed WM+AR compared with calves fed milk replacer where the prevalence in the treatment group declined significantly slower over time. There was no difference between calves fed WM+AR or calves fed milk replacer in the proportion of E. coli isolates that were CTX-M-positive.These findings indicate that feeding WM+AR increased the amount of resistant bacteria shed in the faeces. Shedding of CTX-M-positive E. coli persisted for longer in calves fed WM+AR, and persisted after weaning.

GAPPARD: a computationally efficient method of approximating gap-scale disturbance in vegetation models

Abstract. Models of vegetation dynamics that are designed for application at spatial scales larger than individual forest gaps suffer from several limitations. Typically, either a population average approximation is used that results in unrealistic tree allometry and forest stand structure, or models have a high computational demand because they need to simulate both a series of age-based cohorts and a number of replicate patches to account for stochastic gap-scale disturbances. The detail required by the latter method increases the number of calculations by two to three orders of magnitude compared to the less realistic population average approach. In an effort to increase the efficiency of dynamic vegetation models without sacrificing realism, we developed a new method for simulating stand-replacing disturbances that is both accurate and faster than approaches that use replicate patches. The GAPPARD (approximating GAP model results with a Probabilistic Approach to account for stand Replacing Disturbances) method works by postprocessing the output of deterministic, undisturbed simulations of a cohort-based vegetation model by deriving the distribution of patch ages at any point in time on the basis of a disturbance probability. With this distribution, the expected value of any output variable can be calculated from the output values of the deterministic undisturbed run at the time corresponding to the patch age. To account for temporal changes in model forcing (e.g., as a result of climate change), GAPPARD performs a series of deterministic simulations and interpolates between the results in the postprocessing step. We integrated the GAPPARD method in the vegetation model LPJ-GUESS, and evaluated it in a series of simulations along an altitudinal transect of an inner-Alpine valley. We obtained results very similar to the output of the original LPJ-GUESS model that uses 100 replicate patches, but simulation time was reduced by approximately the factor 10. Our new method is therefore highly suited for rapidly approximating LPJ-GUESS results, and provides the opportunity for future studies over large spatial domains, allows easier parameterization of tree species, faster identification of areas of interesting simulation results, and comparisons with large-scale datasets and results of other forest models.

Mantel–Haenszel estimators of odds ratios for stratified dependent binomial data

A standard approach to analyzing n binary matched pairs usually represented in n 2×2 tables is to apply a subject-specific model; for the simplest situation it is the so-called Rasch model. An alternative population-averaged approach is to apply a marginal model to the single 2×2 table formed by n subjects. For the situation of having an additional stratification variable with K levels forming K 2×2 tables, standard fitting approaches, such as generalized estimating equations and maximum likelihood, or, alternatively, the standard Mantel–Haenszel (MH) estimator, can be applied. However, while all these standard approaches are consistent under a large-stratum limiting model, they are not consistent under a sparse-data limiting model. In this paper, we propose a new MH estimator and a variance estimator that are both dually consistent: consistent under both large-stratum and sparse-data limiting situations. In a simulation study, the properties of the proposed estimators are confirmed, and the estimator is compared with standard marginal methods. The simulation study also considers the case when the homogeneity assumption of the odds ratios does not hold, and the asymptotic limit of the proposed MH estimator under this situation is derived. The results show that the proposed MH estimator is generally better than the standard estimator, and the same can be said about the associated Wald-type confidence intervals.

Accounting for expected attrition in the planning of community intervention trials

Trials in which intact communities are the units of randomization are increasingly being used to evaluate interventions which are more naturally administered at the community level, or when there is a substantial risk of treatment contamination. In this article we focus on the planning of community intervention trials in which k communities (for example, medical practices, worksites, or villages) are to be randomly allocated to each of an intervention and a control group, and fixed cohorts of m individuals enrolled in each community prior to randomization. Formulas to determine k or m may be obtained by adjusting standard sample size formulas to account for the intracluster correlation coefficient rho. In the presence of individual-level attrition however, observed cohort sizes are likely to vary. We show that conventional approaches of accounting for potential attrition, such as dividing standard sample size formulas by the anticipated follow-up rate pi or using the average anticipated cohort size m pi, may, respectively, overestimate or underestimate the required sample size when cluster follow-up rates are highly variable, and m or rho are large. We present new sample size estimation formulas for the comparison of two means or two proportions, which appropriately account for variation among cluster follow-up rates. These formulas are derived by specifying a model for the binary missingness indicators under the population-averaged approach, assuming an exchangeable intracluster correlation coefficient, denoted by tau. To aid in the planning of future trials, we recommend that estimates for tau be reported in published community intervention trials.

Analysis of the individual health benefits accruing from a domestic radon remediation programme

Although radon can be present within buildings at sufficient levels to pose a health risk, levels can be reduced relatively easily. Recent studies on a group of radon-remediated homes, based on assessment of collective population-average risk coefficients, have estimated the benefits and cost effectiveness accruing from remediation and have confirmed that domestic remediation in UK radon Affected Areas would result in significantly reduced cancer risks to the population in those areas. Although the population-average approach used hitherto has applied occupancy and lung-cancer risk factors, these are potentially misleading in assessing discrete populations. The study reported here uses the recently developed European Community Radon Software (ECRS) to quantify individual risks in a sample of householders who remediated their homes following indications that radon levels exceeded the action level. The study proceeds from population-averaged to ‘individual risk’ evaluation, successfully comparing individual and collective risk assessments, and demonstrates that those who remediate are not representative of the general population. Health benefits accruing from remediation are three times lower than expected, largely because remediators are older, live in smaller households, and smoke less than the population average, leading to the conclusion that the current strategy employed in the UK is failing to target those most at risk.

A Comparison of Models for Clustered Binary Outcomes: Analysis of a Designed Immunology Experiment

SUMMARY The lymphocyte proliferative assay (LPA) of immune competence was conducted on 52 subjects, with up to 36 processing conditions per subject, to evaluate whether samples could be shipped or stored overnight, rather than being processed on fresh blood as currently required. The LPA study resulted in clustered binary data, with both cluster level and cluster-varying covariates. Two modelling strategies for the analysis of such clustered binary data are through the cluster-specific and population-averaged approaches. Whereas most research in this area has focused on the analysis of matched pairs data, in many situations, such as the LPA study, cluster sizes are naturally larger. Through considerations of interpretation and efficiency of these models when applied to large clusters, the mixed effect cluster-specific model was selected as most appropriate for the analysis of the LPA data. The model confirmed that the LPA response is significantly impaired in individuals infected with the human immunodeficiency virus (HIV). The LPA response was found to be significantly lower for shipped and overnight samples than for fresh samples, and this effect was significantly stronger among HIV-infected individuals. Surprisingly, an anticoagulant effect was not detected.

NONINFERIORITY TESTING IN CROSSOVER TRIALS WITH CORRELATED BINARY OUTCOMES AND SMALL EVENT PROPORTIONS WITH APPLICATIONS TO THE ANALYSIS OF CONDOM FAILURE DATA

Regulatory agencies may require demonstrating a new therapy or device as noninferior to an existing method. For example, a new condom type must demonstrate noninferiority to existing latex condoms before consideration as an equivalent method of pregnancy prevention. Studies designed to assess condom effectiveness typically measure experimental and standard condom failure in a crossover trial, resulting in unbalanced correlated binary outcomes with low event proportions. We used simulations to evaluate the test size of a simple population average approach to noninferiority testing with the small event proportions, intracluster correlations, and sample sizes frequently found in condom studies. Results emphasize the importance of considering test accuracy when designing any study.

Accommodating negative intracluster correlation with a mixed effects logistic model for bivariate binary data

We extend the random intercept logistic model to accommodate negative intracluster correlations for bivariate binary response data. This approach assumes a single random effect per cluster but entails separate affine transformations of this random effect for the two responses of the pair. We show this approach works for two data sets and a simulation whereas other mixed effects approaches fail. The two data sets are from a crossover trial and a developmental toxicity study of the effects of chemical exposure on malformation risk among rat pups. Comparisons are made with the conditional likelihood approach and with generalized estimating equations estimation of the population-averaged logit model. Simulations show the conditional likelihood approach does not perform well for moderate to strong negative correlations as a positive intracluster correlation is assumed. The proposed mixed effects approach appears to be slightly more conservative than the population-averaged approach with respect to coverage of confidence intervals. Nonetheless, the statistical literature suggests that mixed effects models provide information in addition to that provided by population-averaged models under scientific contexts such as crossover trials. Extensions to trivariate and higher-dimensional responses also are addressed. However, such extensions require certain constraints on the correlation structure.

A Comparison of Cluster-Specific and Population-Averaged Approaches for Analyzing Correlated Binary Data

Summary Clustered or correlated samples of binary responses arise frequently in practice due to repeated measurements or to subsampling the primary sampling units. Several recent approaches address intracluster correlation in binary regression problems including cluster-specific methods such as those based on mixed-effects logistic models and population-averaged methods such as those based on beta-binomial models. This paper considers the interpretations of the regression parameters in these two general approaches. We show that, unlike models for correlated Gaussian outcomes, the parameters of the cluster-specific and population-averaged models for correlated binary data describe different types of effects of the covariates on the response probabilities. In the case of random intercepts, we show that the covariate effects measured by the population-averaged approach are closer to zero than those of the cluster-specific approach when the cluster-specific model holds and that the difference in the magnitude of the covariate effects is increasing with intra-cluster correlation. The case of random slopes is also examined. These results are valid for arbitrary random effects distributions and are demonstrated using data on the ability to obtain samples of breast fluid from women.