Bias Adjustment Research Articles

ASSOCIATION OF MELANOMA AND INCIDENCE OF PARKINSON’S DISEASE: A COHORT STUDY IN US VETERANS

Abstract Melanoma is associated with subsequent Parkinson’s disease (PD) and identifying the clinicopathologic properties of melanoma driving this association could give insight into PD pathophysiology. We undertook a retrospective cohort study of US military Veterans (n=13,737,081) to determine if malignant and premalignant phenotypes were associated with PD by using propensity score matching 4-to-1 and adjustment for competing risk of death and key confounders. Broad cancer categories were obtained from the Oncology VA registry. Data about melanoma were obtained from clinical and pathology reports using manually validated natural language processing. Risk difference and 95% confidence intervals (CI) were estimated by presence or absence of cancer. For broad cancer categories, we found decreased PD incidence associated with multiple malignancies with some novel associations (e.g. HR = 0.4 [0.2,0.5] for myeloma). In terms of melanoma, BRAF and NRAS mutational status was not associated with incident PD. Atypical/dysplastic nevi was associated with increased PD incidence (HR = 1.4 [1.2,1.7]). Melanoma subtypes in the Oncology registry were associated with PD and included superficial spreading type (HR = 1.8 [1.0, 3.3]) and lentigo maligna (HR = 1.6 [1.2, 2.1]). In people with VA pathology reports, this effect was replicated with greater effect size for superficial spreading type (HR = 10.4 [6.5,16.9]) and lentigo maligna (HR = 6.4 [3.4,11.8]) although adjustment for ascertainment bias is needed. These data show PD is associated with melanoma subtypes and precursor lesions and suggest screening for PD could be appropriate in those with melanocytic lesions once early diagnostic tests are developed.

Elucidating postoperative dynamics in tractional retinal detachment: a systematic review and meta-analysis of structural and functional outcomes following diabetic vitrectomy, including an analysis of postoperative complications.

This meta-analysis aims to evaluate the efficacy and safety of diabetic vitrectomy for treating tractional retinal detachment (TRD) by assessing visual acuity improvements, retinal reattachment success rates, and the frequency of postoperative complications. A comprehensive literature search identified 1,211 studies, 30 of which met the inclusion criteria and were analyzed. The meta-analyses were conducted via random effects models to account for heterogeneity. The primary outcomes were visual acuity improvement and retinal reattachment success, whereas postoperative complications were analyzed as secondary outcomes. Funnel plots, trim-and-fill methods, and Egger's regression tests were employed to assess and adjust for publication bias. The meta-analysis included data from 1,844 eyes across 30 studies. The pooled mean difference (MD) in visual acuity following diabetic vitrectomy was 0.80 logMAR (95% CI: 0.58-1.01, p < 0.0001), with significant heterogeneity (I² = 95%). The trim-and-fill method adjusted the MD to 1.15 (95% CI: 0.85-1.45), indicating potential publication bias. The pooled proportion of retinal reattachment success was 94.63% (95% CI: 90.88-96.89%), with substantial heterogeneity (I² = 82.2%). Egger's regression test confirmed publication bias in the retinal reattachment data (t = 6.07, df = 25, p < 0.0001). The most common postoperative complications were vitreous hemorrhage (8.2%, 0-41%), elevated intraocular pressure (7.3%, 0-25%), and cataract formation (10.3%, 0-43%). Diabetic vitrectomy significantly improves visual acuity and achieves high retinal reattachment success rates in patients with TRD despite considerable heterogeneity across studies. However, a critical limitation of the current literature is the lack of comprehensive reporting of essential markers of perfusion, such as the choroidal vascularity index and choriocapillaris flow area, which are crucial for a thorough understanding of the impact of vitrectomy on ocular perfusion and its correlation with structural and functional outcomes. These findings also highlight the need for vigilant postoperative management to mitigate common complications. Adjustments for publication bias via the trim-and-fill method suggest that the initial estimates of efficacy may be optimistic, but the overall benefits of the procedure remain substantial. Further research is warranted to standardize surgical techniques, include detailed perfusion outcomes, and enhance the reliability of future meta-analyses. Trial registration: Retrospectively registered.

A Systematic Review of Implicit Versus Explicit Social Skills Group Programs in Different Settings for School-Aged Autistic Children and Adolescents.

This systematic review and meta-analysis aimed to identify and evaluate the quality of randomised controlled trials (RCTs), assessing the efficacy of Social Skills Group Programs (SSGPs) for primary and secondary school aged autistic children and adolescents exploring the influence of informant, setting characteristics and teaching strategies as moderating factors for social outcomes. A search of the electronic databases of Medline, ProQuest, CINAHL, Scopus, and Web of Sciences electronic databases was conducted for the period January 1, 2013 until December 31, 2023 for peer-reviewed RCT studies published in English, evaluating the efficacy of SSGPs for school-aged autistic children and adolescents (6-18years). Included studies were assessed for methodological quality and efficacy using random effect meta-analysis. Informant analysis and moderator analyses were also conducted investigating the influence of informant, setting characteristics and teaching strategy utilised in the SSGPs on the participants social outcomes. Sixty-five studies were included in the systematic review with 49 studies included in the meta-analysis. Although most studies had strong or good methodological quality, a high possibility of publication bias was detected in the meta-analysis. After statistical adjustments for publication bias were made, findings revealed that SSGPs had no effect on the overall outcomes assessed by included studies on school-aged autistic children and adolescents. However, informant analysis revealed small effects reported by self-report, parent-proxy and researchers. Teaching strategies and setting characteristics were not significant moderators for the efficacy of SSGPs on the social outcomes of autistic children and adolescents. This review highlights the need for improvements in measurement frameworks for assessing social skills in autistic children and adolescents.

Impacts of LULC and climate change on runoff and sediment production for the Puyango-Tumbes basin (Ecuador-Peru)

Climate change will cause alterations in the hydrological cycle, a topic of great relevance to the scientific community due to its impacts on water resources. Investigating changes in hydrological characteristics at the watershed level in the context of climate change is fundamental for developing mitigation and adaptation strategies against extreme hydrological events. This study aimed to analyze the impacts of climate change on flow and sediment production in the Puyango-Tumbes watershed. Projected climate data from CMIP6 were used, corrected through a bias adjustment process to minimize discrepancies between model data and historical observations, ensuring a more accurate representation of climate behavior. The analysis combined two representative climate change scenarios (SSP2-4.5 and SSP5-8.5) with two land use and land cover (LULC) scenarios: (a) an optimistic scenario with reduced anthropogenic effects (LULC_1985) and (b) a pessimistic scenario reflecting future impacts (LULC_2015). The SWAT model estimated future flow and sediment production for two periods (2035-2065 and 2070-2100), following model calibration and validation against the reference period 1981-2015 at three hydrometric stations: Pindo, Puyango, and El Tigre, located in Ecuador and Peru. The simulations revealed a significant increase in sediment generation under the pessimistic scenario SSP5-8.5, followed by SSP2-4.5, while lower sediment yields were observed in the optimistic scenarios. Even in the best-case scenario (optimistic SSP2-4.5), sediment yields remained substantially higher than the reference conditions. Additionally, higher flows were anticipated in some scenarios, with the El Tigre station in the lower watershed being the most affected area. These findings underscore the high probability of more frequent flooding events due to increased sediment yields and flow variability. The results highlight the urgent need for implementing adaptation measures, such as improved land use management and hydrological infrastructure, to enhance social resilience and mitigate the impacts of climate change in the watershed.

Survival Analysis of Conversion Surgery in Borderline Resectable and Locally Advanced Unresectable Pancreatic Ductal Adenocarcinoma Addressing Selection and Immortal Time Bias: A Retrospective Single-Center Study.

The purpose of this study was to provide a detailed evaluation of the oncological advantages of surgery following neoadjuvant chemotherapy (NAC) for patients with borderline resectable (BR) or unresectable (UR) pancreatic ductal adenocarcinoma (PDAC), with a focus on minimizing biases. Recently, NAC has become the standard care for BR or UR locally advanced (UR-LA) PDAC, however, many studies have assessed survival benefits and favorable variables without consideration for biases, particularly immortal time bias. This study included patients diagnosed with BR or UR-LA PDAC at Juntendo University Hospital from 2019 to 2022. To mitigate bias, we applied methods such as propensity score matching (PSM), time-dependent covariate Cox proportional hazard regression analysis (TDC), landmark analysis, and multivariable Cox proportional hazards regression model. The study analyzed 124 patients, dividing them into a surgery group (n = 57) and a chemotherapy-only group (n = 67). After PSM, there were 21 matched pairs. Survival analysis using TDC analysis showed that the surgery group had significantly better overall survival compared with the chemotherapy-only group in both the entire cohort and the matched pairs. Cox regression analysis of the entire cohort also revealed a similar superiority of surgery, while the landmark analysis showed varying results depending on the landmark setting. After careful adjustment for selection and immortal time biases, surgery following NAC appears to significantly extend survival in patients with BR or UR PDAC.

Thirty-Day Mortality of Persistent Methicillin-Resistant Staphylococcus aureus Bacteremia: Insights from a Retrospective Cohort Study.

Although glycopeptides remain the preferred treatment for methicillin-resistant Staphylococcus aureus (MRSA) bacteremia, the treatment of persistent MRSA bacteremia has been challenging. We investigated real-world treatment strategies for persistent MRSA bacteremia, with a specific emphasis on the use of antimicrobial agents and the frequency of changes during the treatment course. We retrospectively identified patients with persistent MRSA bacteremia in four university-affiliated hospitals between 2017 and 2021. The primary objective of this study was to investigate the patterns of antimicrobial uses for MRSA bacteremia. The secondary objectives were evaluating the associated factors with 1) overall 30-day mortality and 2) changing agents during the treatment course. Time-dependent Cox regression analysis was used to adjust for immortal time bias. Among 116 patients, 37.1% underwent antimicrobials switching, primarily prompted by persistent bacteremia. The 30-day mortality rates of groups with and without antimicrobial switching were 21.4% and 44.2%, respectively (p=0.010 by log-rank test); however, after adjustment for immortal time bias, there was no statistical significance between the two groups (adjusted hazard ratio 0.24, 95% confidence interval 0.03-2.17, p=0.238). Only the Pitt bacteremia score on day 4 and pneumonia were associated with 30-day mortality. Meanwhile, the factors associated with antimicrobial switching were the duration of bacteremia, the initial use of teicoplanin, echocardiogram, and Charlson comorbidity index. This study showed that while over one-third of persistent MRSA bacteremia patients experience changes in antimicrobial agents during treatment, this practice does not significantly improve the 30-day mortality. Our study suggests the need for more effective treatment strategies in managing persistent MRSA bacteremia.

GALILEO HAS overview, comparison with analogues, and assessment of perspective

In 2022, Galileo, the European GNSS, launched the first phase of its HAS (High Accuracy Service) initiative. With free and global corrections for clock delays and satellite orbits, Galileo HAS provides decimeter positioning accuracy without additional ground networks. This research aims to evaluate the advancements in precise positioning technologies, focusing on the Galileo High Accuracy Service (HAS). The study highlights the importance of precision positioning methods, including Standard Point Positioning (SPP), Real-Time Kinematic (RTK), and Precise Point Positioning (PPP). It assesses the performance of Galileo HAS in comparison with other global augmentation services like QZSS CLAS and BeiDou PPP-B2b. The methodology involves a comprehensive analysis of the technical capabilities, accuracy, and operational limitations of HAS through a literature review and comparative analysis of positioning performance data. The results confirm that Galileo HAS achieves decimeter-level accuracy globally, with horizontal accuracy below 20 cm and vertical accuracy below 40 cm. However, it suffers from prolonged convergence times due to the absence of atmospheric and phase bias corrections in its initial phase. The scientific novelty lies in identifying HAS's potential as the first global free PPP correction service via Signal-in-Space (SIS), distinguishing its practical advantages in semi-enclosed environments compared to traditional PPP augmentation systems. The study also emphasizes the unique integration challenges posed by HAS corrections due to proprietary encoding formats. Thee findings underscore HAS's utility in geodesy, mapping, and real-time applications, particularly in resource-constrained settings. However, the research highlights critical areas for improvement, such as implementing atmospheric corrections and phase bias adjustments, to meet real-time precise positioning demands. The conclusions note the undoubted usefulness of such a service as Galileo HAS and review its shortcomings and methods of solving them.

A consistent ocean oxygen profile dataset with new quality control and bias assessment

Abstract. Global ocean oxygen concentrations have declined in the past decades, posing threats to marine life and human society. High-quality and bias-free observations are crucial to understanding ocean oxygen changes and assessing their impact. Here, we propose a new automated quality control (QC) procedure for ocean profile oxygen data. This procedure consists of a suite of 10 quality checks, with outlier rejection thresholds being defined based on underlying statistics of the data. The procedure is applied to three main instrumentation types: bottle casts, CTD (conductivity–temperature–depth) casts, and Argo profiling floats. Application of the quality control procedure to several manually quality-controlled datasets of good quality suggests the ability of the scheme to successfully identify outliers in the data. Collocated quality-controlled oxygen profiles obtained by means of the Winkler titration method are used as unbiased references to estimate possible residual biases in the oxygen sensor data. The residual bias is found to be negligible for electrochemical sensors typically used on CTD casts. We explain this as the consequence of adjusting to the concurrent sample Winkler data. Our analysis finds a prevailing negative residual bias with the magnitude of several µmol kg−1 for the delayed-mode quality-controlled and adjusted profiles from Argo floats varying among the data subsets adjusted by different Argo Data Assembly Centers (DACs). The respective overall DAC- and sensor-specific corrections are suggested. We also find the bias dependence on pressure, a feature common to both AANDERAA optodes and SBE43-series sensors. Applying the new QC procedure and bias adjustments resulted in a new global ocean oxygen dataset from 1920 to 2023 with consistent data quality across bottle samples, CTD casts, and Argo floats. The adjusted Argo profile data are available at the Marine Science Data Center of the Chinese Academy of Sciences (https://doi.org/10.12157/IOCAS.20231208.001, Gouretski et al., 2024).

Parameterization of Beta Distributions for Bias Parameters of Binary Exposure Misclassification in Probabilistic Bias Analysis.

To account for misclassification of dichotomous variables using probabilistic bias analysis, beta distributions are often assigned to bias parameters (e.g., positive and negative predictive values) based on data from an internal validation substudy. Due to the small sample size of validation substudies, zero-cell frequencies can occur. In these scenarios, it may be helpful to assign prior distributions or apply continuity corrections to the predictive value estimates. We simulated cohort studies of varying sizes, with a binary exposure and outcome and a true risk ratio (RR) = 2.0, as well as internal validation substudies, to account for exposure misclassification. We conducted bias adjustment under five approaches assigning prior distributions to the positive and negative predictive value parameters: (1) conventional method (i.e., no prior), (2) uniform prior beta ( α = 1, β = 1), (3) Jeffreys prior beta ( α = 0.5, β = 0.5), (4) using Jeffreys prior as a continuity correction only when zero cells occurred, and (5) using the uniform prior as a continuity correction only when zero cells occurred. We evaluated performance by measuring coverage probability, bias, and mean squared error. For sparse validation data, methods (2)-(5) all had better coverage and lower mean squared error than the conventional method, with the uniform prior (2) yielding the best performance. However, little difference between methods was observed when the validation substudy did not contain zero cells. If sparse data are expected in a validation substudy, using a uniform prior for the beta distribution of bias parameters can improve the validity of bias-adjusted measures.

Robust handling of extremes in quantile mapping – “Murder your darlings”

Abstract. Quantile mapping is a method often used for the bias adjustment of climate model data toward a reference, i.e. to construct a transformation of the model's distribution to that of the reference. The main moments of the distributions are typically well transformed by quantile mapping, but statistical uncertainty increases towards the extreme tails, making robust transformations challenging. Because of the limited data at the extreme tails, an empirical quantile mapping also needs to make some estimation or fit a parameterised function for data beyond the calibration data range. Here, the MIdAS bias adjustment platform is employed to explore different methods for handling the extreme tail; these approaches are evaluated using an indicator of extreme precipitation – the maximum daily precipitation amount per year. Different methodologies are evaluated for a large ensemble of regional climate model projections over Scandinavia. The sensitivity of the empirical quantile mapping to the tails of the distribution is demonstrated, and it is found that the behaviour is significantly different within and outside of the calibration period, causing severe issues with the temporal consistency of the time series. The sensitivity is identified to be due to differences in the activated features of the bias adjustment within the calibration period (where the empirical transfer function is applied) and outside of that period (where the extrapolation method is likely applied). This means that the bias adjustment method is, in a sense, different between different time periods. Furthermore, finding a robust parameterisation for the tail is not straightforward. We identify a two-step solution that works well for this problem: We refer to the first step as “Murder your darlings”. By excluding data from the tail data in the calibration period, the extrapolation feature is activated for all time periods, even the calibration period. In the second step, applying an outlier-insensitive method for linear regression works well for finding an extrapolation parameterisation for the tail.

Improved wind power assessments by bias adjusted reanalysed data with applications near Morocco’s coast

The wind energy requirements for high temporal and spatial observations, unavailable in many countries, are met by reanalysed wind data, which are increasingly used in wind power assessments. However, these data also include unrealistic features, making their validation compulsory. The validation analysis of ERA5 and MERRA-2 wind data, two of the most successful wind reanalyses, with observations near Morocco’s coast, highlighted the existence of bias in both datasets, and other shortcomings. The study was conducted at eight selected sites in the period 2011-2020 using linear and probability density function based statistical tests. The local ERA5-BA and MERRA-2 BA velocity fields were created using Bias Adjustment (BA) Techniques. Only part of the observed and modelled datasets (training period) were used to identify the transfer function. The BA fields were verified against observations during the remaining period (verification period). They outperformed their respective non-adjusted fields and also the WFDE5 (the global bias-corrected ERA5 wind field) in the wind potential assessments, represented by capacity factor and low-wind day. At every location, the relative errors in the latter’s estimates using ERA5-BA were lowered to less than 10 %, and the same was true for MERRA-2-BA. Thus, an observational dataset of reduced length can help overcome some reanalysed wind data limitations for wind power estimations. Additionally, this methodology can be applied to the correction of each reanalysis update.

Nonparticipation in a Digital Health Intervention Study Among Older Adults: Uneven Involvement, Biased Outcomes, and the Effect of Weighting.

The involvement of older adults in research on digital health is uneven with respect to, for example, age, gender, health status, and digital skills. However, little is known regarding the effect of the uneven involvement of older adults in digital health research on researched outcomes. This study helps to fill this knowledge gap, identifies the effects of uneven involvement of older adults in digital health research on researched outcomes, and also develops a correction for this. Data are retrieved from a digital health intervention for postoperative monitoring of people who underwent day surgery in Sweden. Based on field information on the recruitment process and researched outcomes for the intervention, this study (i) tested intervention effects by using 2 standard unweighted procedures in a sample of 281 individuals aged 50 years or older, and then (ii) used the information on participants, nonparticipants, and their respective probabilities to be involved in the intervention study to perform a weighting of the intervention effects for each step of selection and for the study group membership. The intervention effects were found to be overestimated due to overrepresentation of groups that gained from receiving the intervention. No intervention effects were found after adjustment for participation bias. Selective participation of older adults in digital health research biases research outcomes and can lead to overestimation of intervention effects. Weighting allows researchers to correct and describe the effect of selective participation on researched outcomes.

Quantifying Shape and Texture Biases for Enhancing Transfer Learning in Convolutional Neural Networks

Neural networks have inductive biases owing to the assumptions associated with the selected learning algorithm, datasets, and network structure. Specifically, convolutional neural networks (CNNs) are known for their tendency to exhibit textural biases. This bias is closely related to image classification accuracy. Aligning the model’s bias with the dataset’s bias can significantly enhance performance in transfer learning, leading to more efficient learning. This study aims to quantitatively demonstrate that increasing shape bias within the network by varying kernel sizes and dilation rates improves accuracy on shape-dominant data and enables efficient learning with less data. Furthermore, we propose a novel method for quantitatively evaluating the balance between texture bias and shape bias. This method enables efficient learning by aligning the biases of the transfer learning dataset with those of the model. Systematically adjusting these biases allows CNNs to better fit data with specific biases. Compared to the original model, an accuracy improvement of up to 9.9% was observed. Our findings underscore the critical role of bias adjustment in CNN design, contributing to developing more efficient and effective image classification models.

Changes in Reader Performance During Sequential Reading of Breast Cancer Screening Digital Breast Tomosynthesis Examinations.

Background Studies suggest that readers experience perceptual adaptation when interpreting batched screening mammograms, which may serve as a mechanism for improved performance. Purpose To analyze clinical digital breast tomosynthesis (DBT) screening data to evaluate changes in reader performance during sequential batch reading. Materials and Methods This observational retrospective study used data from the radiology information system collected for screening DBT examinations performed from January 2018 to December 2019. The reference standard was established based on pathology results, if applicable, or imaging findings at 1-year follow-up. Examinations were aggregated into batches, defined as a sequence of cases for a given reader with differences in interexamination time of 10 minutes or less. Mixed-effect models were used to evaluate performance and timing across batch positions. Statistical adjustments accounted for potential confounders, including patient characteristics (age, breast density), reading factors (day of week, time of day), and between-reader heterogeneity. Results The dataset included 121 652 examinations (median patient age, 61 years [IQR, 53-69 years]), including 1081 cancers interpreted by 15 radiologists. Unadjusted false-positive rates decreased from an averaged 15.5% at the first within-batch examination to 10.5% after three sequentially read examinations (P < .001), without a significant change in sensitivity (82.6% vs 84.2%; P = .15). The interpretation time consistently decreased, achieving a substantial reduction in longer batches (average 2.8 to 2.2 minutes for noncancer examinations; P < .001). Adjustment for sampling bias and confounders (patient age, breast density, day of week, time of day; all P < .005) reduced the effect on false-positive rates (11.5% to 9.4% after three examinations; P < .001) and interpretation time (3.2 to 2.7 minutes for noncancer examinations; P < .001). Conclusion Radiologists showed improved recall rates and interpretation times without change in sensitivity over the course of batch reading of DBT screening examinations. © RSNA, 2024 See also the editorial by Iima and Satake in this issue.

Abstract 245: Dose‐dependent relationship between periodontal disease and ischemic stroke

Introduction The link between periodontal disease and cerebrovascular disease was reported in various studies, with several mechanistic underpinnings. To determine the predictive value of periodontal disease for cerebrovascular diseases, we aimed to investigate the quantitative association between various degrees of periodontal disease and subtypes of cerebrovascular disease. Methods Using Meta‐analyses Of Observational Studies in Epidemiology and PRISMA guidelines, we searched 6 databases through May 2024 for cohort and case‐control studies evaluating the relationship between periodontal disease and ischemic or hemorrhagic stroke. Studies were included if they reported risk ratios (RRs) and 95% confidence intervals or raw patient‐level data (that were computed into RRs). Unadjusted and vascular risk‐factor adjusted ORs were pooled into meta‐analysis using DerSimonian Laird random effects model. Study quality, publication and dissemination biases were assessed and integrated in the analyses (RevMan software). Results 12 prospective studies encompassing 31, 324 patients were eligible for the qualitative and quantitative syntheses. After vascular risk‐factor adjustment, any degree of periodontitis was associated with an increased risk of ischemic stroke (RR, 2.24, 95% CI [1.67‐2.98], p&lt;0.001, I2=10). Subgroup analysis including subjects with moderate or severe periodontitis showed a higher risk of ischemic stroke (RR, 3.01, 95% CI [1.01‐4.05], p&lt;0.001). Associations remained significant after adjustments for quality and publication biases. Heterogeneity was low (I2&lt;20%) for all syntheses. The reported mechanisms were cardioembolism and thromboembolism. The risk of hemorrhagic stroke could not be synthesized due to limited data. Conclusions A dose‐dependent relationship exists between periodontal disease and ischemic stroke, even after adjustment for traditional cerebrovascular risk factors. It remains unknown if the risk of hemorrhagic stroke is increased in subjects with periodontal disease. Periodontal disease prevention and treatment may decrease the incidence of ischemic cerebrovascular disease.

A non-stationary bias adjustment method for improving the inter-annual variability and persistence of projected precipitation

Hydrological studies depend heavily on environmental variables, such as precipitation and resulting runoff, which exhibit highly seasonal and intermittent behaviour in semiarid basins. In these basins, the use of traditional methods to adjust biases in time series projections can lead to inaccurate results regarding the impacts of climate change. This study introduces a non-stationary bias adjustment methodology (NS) specifically designed for environmental variables characterized by sporadic events and substantial intensity variability, such as precipitation. The methodology involves establishing a probability threshold to adapt the occurrence of precipitation events and utilizes a quantile mapping method based on a non-stationary theoretical and parametric distribution to adjust biases associated with precipitation intensity. The NS method is applied to daily precipitation projections from seven regional climatic models under the RCP 8.5 scenario spanning 2006–2100, alongside historical concurrent data from 1970 to 2005. The present method is compared to the widely used quantile delta mapping approach (QDM), revealing significant differences in performance related to the distribution of precipitation events throughout the year and the behaviour of mean and extreme intensity values. Both approaches show reliable performance, with root mean square errors between the empirical distribution functions of the corrected hindcast time series and the observations being lower than or close to 1 mm for the percentiles smaller than 50th. The error increases with the percentiles, particularly at stations located at higher altitudes. In these locations, QDM doubles or triples the error obtained with the non-stationary approach for percentiles higher than 75th, reaching up to 34 mm. The proposed methodology demonstrates promising potential in reducing uncertainties associated with systematic errors in inter-annual precipitation variability. It is a first step to jointly apply a similar approach to all involved driving variables. Such methods are key to assessing hydrological responses and associated impacts in semi-arid mountainous basins all around the Globe.

Bias analysis on socioeconomic status and lung cancer in a pooled international case-control study

Abstract Background Low socioeconomic status (SES) groups showed increased lung cancer risks even after adjustment for smoking habits and other exposures to lung carcinogens. Although several biases were often indicated and discussed, only few studies quantified the impact of potential biases. Methods We conducted a bias analysis on the association of lung cancer and SES using data from the SYNERGY project, including 12 case-control studies with 18 study centres from Europe and Canada (16,550 cases, 20,147 controls). SES in quartiles was derived from the International Socio-Economic Index of occupational status (ISEI). Odds ratios (OR) with 95% confidence intervals (CI) were estimated by logistic regression adjusting for age, study centre, and smoking. In addition, we estimated natural direct SES effects and natural indirect smoking effects by inverse odds ratio weighting. In a multiple quantitative bias analysis, we considered impacts of misclassification of smoking status, selection bias, and unmeasured mediator-outcome confounding by a protective genetic factor, and created 95% simulation intervals (SI) by bootstrap. All analyses were stratified by sex. Results Adjustment for smoking as well as natural effects estimation showed that nearly half of lung cancer risks of lower SES groups in men and up to one third in women were attributable to smoking. Consideration of all types of bias reduced lung-cancer risks in the fully adjusted logistic regression models, with the strongest impact by selection bias: For the 4th versus 1st (highest) ISEI quartile OR decreased from 1.83 (1.69-1.98 CI) to 1.50 (1.30-1.73 SI) in men, and OR 1.48 (1.27-1.72 CI) to 1.20 (0.96-1.53 SI) in women. Conclusions Smoking is the main target for prevention of lung cancer, along with occupational and environmental exposures, in particular in lower SES groups. This finding remains, though our analysis revealed reduced lung-cancer risks of lower SES groups after multiple bias adjustment. Key messages • Impact of potential biases was quantified in the association of SES and lung cancer. • Lung cancer risks were partially attributable to smoking and multiple biases.

A Forensic Investigation of Climate Model Biases in Teleconnections: The Case of the Relationship Between ENSO and the Northern Stratospheric Polar Vortex

AbstractTeleconnections are crucial in shaping climate variability and regional climate change. The fidelity of teleconnections in climate models is important for reliable climate projections. As the observed sample size is limited, scientific judgment is required when models disagree with observed teleconnections. We illustrate this using the example of the relationship between El Niño‐Southern Oscillation (ENSO) and the northern stratospheric polar vortex (SPV), where the MIROC6 large ensemble exhibits an ENSO‐SPV correlation opposite in sign to observations. Yet the model well captures the upward planetary‐wave propagation pathway through which ENSO is known to affect the SPV. We show that the discrepancy arises from the model showing an additional linkage related to horizontal stratospheric wave propagation. Observations do not provide strong statistical evidence for or against the existence of this linkage. Thus, depending on the research purpose, a choice has to be made in how to use the model simulations. Under the assumption that the additional linkage is spurious, a physically‐based bias adjustment is applied to the SPV, which effectively aligns the modeled ENSO‐SPV relationship with the observations, and thereby removes the model‐observations discrepancy in the surface air temperature response. However, if one believed that the additional linkage was genuine and was undersampled in the observations, a different approach could be taken. Our study emphasizes that caution is needed when concluding that a model is not suitable for studying teleconnections. We propose a forensic approach and argue that it helps to better understand model performance and utilize climate model data more effectively.

How Does Survey Timing Influence Apparent Wasting Trends? A Case Study from Senegal

BackgroundChild wasting is known to exhibit seasonal patterns, but few studies have examined how the seasonality of wasting affects tracking of wasting trends over multiyear periods. ObjectivesWe explored the seasonality of wasting in Senegal relative to multiyear changes and examined implications for tracking. We tested whether month-fixed effects reduced bias in estimating longer-term wasting trends given variation in survey timing. MethodsThe average prevalence of child wasting (weight-for-height z-score < −2) and 95% confidence intervals were calculated by month and year from the continuous Demographic and Health Surveys (DHS) from 2012–2019. Peak and low wasting season estimates were defined as the 4 highest and the 4 lowest months of average wasting prevalence. Month-adjusted annual wasting estimates were generated using month-fixed effects linear regression, and the effectiveness of this method of bias adjustment was examined in simulated datasets. ResultsNationally, wasting fluctuated from 2013–2019, with the lowest annual prevalence of 6.0% (95% confidence interval [CI]: 5.1, 7.0%) recorded in 2014 and the highest in 2017 at 9.0% (95% CI: 8.3, 9.8%). Pooled across years, the peak wasting season prevalence was 8.8% (95% CI: 8.3, 9.3%), and low wasting season prevalence was 6.4% (95% CI: 5.7, 7.1%). Month-adjusted wasting estimates did not differ notably from raw wasting prevalence estimates. Simulations demonstrated that adjusting for months reduces bias in wasting when surveys are conducted 1 or 2 mo apart across waves but fails to reliably do so when surveys are conducted in different seasons across waves. ConclusionsSeasonal fluctuations in the prevalence of wasting can be large enough to bias the interpretation of multiyear trends. Efforts should be made to conduct national surveys at the same time of year. Seasonality adjustment using month-fixed effects works more reliably when the differences in survey periods across waves are minimal.

Downscaling, bias correction, and spatial adjustment of extreme tropical cyclone rainfall in ERA5 using deep learning

Hydrological models that are used to analyse flood risk induced by tropical cyclones often input ERA5 reanalysis data. However, ERA5 precipitation has large systematic biases, especially over heavy precipitation events like Tropical Cyclones, compromising its usefulness in such scenarios. Few studies to date have performed bias correction of ERA5 precipitation and none of them for extreme rainfall induced by tropical cyclones. Additionally, most existing works on bias adjustment focus on adjusting pixel-wise metrics of bias, such as the Mean Squared Error (MSE). However, it is equally important to ensure that the rainfall peaks are correctly located within the rainfall maps, especially if these maps are then used as input to hydrological models. In this paper, we describe a novel machine learning model that addresses both gaps, RA-Ucmpd, based on the popular U-Net model. The key novelty of RA-Ucmpd is its loss function, the compound loss, which optimizes both a pixel-wise bias metric (the MSE) and a spatial verification metric (a modified version of the Fractions Skill Score). Our results show how RA-Ucmpd improves ERA5 in almost all metrics by 3-28%—more than the other models we used for comparison which actually worsen the total rainfall bias of ERA5—at the cost of a slightly increased (3%) error on the magnitude of the peak. We analyse the behaviour of RA-Ucmpd by visualizing accumulated maps of four particularly wet tropical cyclones and by dividing our data according to the Saffir-Simpson scale and to whether they made landfall, and we perform an error analysis to understand under what conditions our model performs best.