BMC Medical Research Methodology最新文献

Objective: Dementia is a significant medical and social issue in most developed countries. Practical tools for predicting the progression of degenerative dementia are highly valuable. Machine learning (ML) methods facilitate the construction of effective models using real-world data, which may include missing values and various integrated datasets.

Method: This retrospective study analyzed data from 679 patients diagnosed with degenerative dementia at Fu Jen Catholic University Hospital, who were evaluated by neurologists, psychologists and followed for over two years. Predictive variables were categorized into demographic (D), clinical dementia rating (CDR), mini-mental state examination (MMSE), and laboratory data value (LV) groups. These categories were further integrated into three subgroups (D-CDR, D-CDR-MMSE, and D-CDR-MMSE-LV). We utilized the extreme gradient boosting (XGB) model to rank the importance of variables and identify the most effective feature combination via a step-wise approach.

Result: The D-CDR-MMSE-LV model combination showed robust performance with an excellent area under the receiver operating characteristic curve (AUC) and the highest sensitivity value (84.66). Employing both demographic and neuropsychiatric variables, our prediction model achieved an AUC of 83.74. By incorporating additional clinical information from laboratory data and applying our proposed feature selection strategy, we constructed a model based on eight variables that achieved an AUC of 85.12 using the XGB technique.

Conclusion: We established a machine-learning model to monitor the progression of dementia using a limited, real-world clinical dataset. The XGB technique identified eight critical variables across our integrated datasets, potentially providing clinicians with valuable guidance.

Background: Rapid innovation and new regulations lead to an increased need for post-marketing surveillance of implantable devices. However, complex multi-level confounding related not only to patient-level but also to surgeon or hospital covariates hampers observational studies of risks and benefits. We conducted parametric and plasmode simulations to compare the performance of cardinality matching (CM) vs propensity score matching (PSM) to reduce confounding bias in the presence of cluster-level confounding.

Methods: Two Monte Carlo simulation studies were carried out: 1) Parametric simulations (1,000 iterations) with patients nested in clusters (ratio 10:1, 50:1, 100:1, 200:1, 500:1) and sample size n = 10,000 were conducted with patient and cluster level confounders; 2) Plasmode simulations generated from a cohort of 9981 patients admitted for pancreatectomy between 2015 to 2019 from a US hospital database. CM with 0.1 standardised mean different constraint threshold (SMD) for confounders and PSM were used to balance the confounders for within-cluster and cross-cluster matching. Treatment effects were then estimated using logistic regression as the outcome model on the obtained matched sample.

Results: CM yielded higher sample retention but more bias than PSM for cross-cluster matching in most scenarios. For instance, with ratio of 100:1, sample retention and relative bias were 97.1% and 26.5% for CM, compared to 82.5% and 12.2% for PSM. The results for plasmode simulation were similar.

Conclusions: CM offered better sample retention but higher bias in most scenarios compared to PSM. More research is needed to guide the use of CM particularly in constraint setting for confounders for medical device and surgical epidemiology.

Background: The concept of the population is of fundamental importance in epidemiology and statistics. In some instances, it is not possible to sample directly from the population of interest. Weighting is an established statistical approach for making inferences when the sample is not representative of this population.

Methods: The effective sample size (ESS) is a descriptive statistic that can be used to accompany this type of weighted statistical analysis. The ESS is an estimate of the sample size required by an unweighted sample that achieves the same level of precision as the weighted sample. The ESS therefore reflects the amount of information retained after weighting the data and is an intuitively appealing quantity to interpret, for example by those with little or no statistical training.

Results: The conventional formula for calculating ESS is derived under strong assumptions, for example that outcome data are homoscedastic. This is not always true in practice, for example for survival data. We propose three new approaches to compute the ESS, that are valid for any type of data and weighted statistical analysis, and so can be applied more generally.

Conclusion: We illustrate all methods using an example and conclude that our proposals should accompany, and potentially replace, the existing approach for computing the ESS.

Data literacy, the ability to understand and effectively communicate with data, is crucial for researchers to interpret and validate data. However, low reproducibility in biomedical research is nowadays a significant issue, with major implications for scientific progress and the reliability of findings. Recognizing this, funding bodies such as the European Commission emphasize the importance of regular data management practices to enhance reproducibility. Establishing a standardized framework for statistical methods and data analysis is essential to minimize biases and inaccuracies. The FAIR principles (Findable, Accessible, Interoperable, Reusable) aim to enhance data interoperability and reusability, promoting transparent and ethical data practices. The study presented here aimed to train postgraduate students at the Universidad Europea de Madrid in data literacy skills and FAIR principles, assessing their application in master thesis projects. A total of 46 participants, including students and mentors, were involved in the study during the 2022-2023 academic year. Students were trained to prioritize FAIR data sources and implement Data Management Plans (DMPs) during their master's thesis. An 11-item questionnaire was developed to evaluate the FAIRness of research data, showing strong internal consistency. The study found that integrating FAIR principles into educational curricula is crucial for enhancing research reproducibility and transparency. This approach equips future researchers with essential skills for navigating a data-driven scientific environment and contributes to advancing scientific knowledge.

Background: The use of the multistate Markov chain model is a valuable tool for studying child undernutrition. This allows us to examine the trends of children's transitions from one state to multiple states of undernutrition.

Objectives: In this study, our objective was to estimate the median duration for a child to first transition from one state of undernutrition to another as well as their first recurrence of undernutrition and also to analyze the typical duration of undernourishment. This involves understanding the central tendency of these transitions and durations in the context of longitudinal data.

Methods: We used a longitudinal dataset from the Young Lives cohort study (YLCS), which included approximately 1997 Ethiopian children aged 1-15 years. These children were selected from five regions and followed through five survey rounds between 2002 and 2016. The surveys provide comprehensive health and nutrition data and are designed to assess childhood poverty. To analyze this dataset, we employed a Markov chain regression model. The dataset constitutes a cohort with repeated measurements, allowing us to track the transitions of individual children across different states of undernutrition over time.

Results: The findings of our study indicate that 46% of children experienced concurrent underweight, stunting, and wasting (referred to as USW). The prevalence of underweight and stunted concurrent condition (US) was 18.7% at baseline, higher among males. The incidence density of undernutrition was calculated at 22.5% per year. On average, it took 3.02 months for a child in a wasting state to transition back to a normal state for the first time, followed by approximately 3.05 months for stunting and 3.89 months for underweight. It is noteworthy that the median duration of undernourishment among children in the US (underweight and stunted concurrently) state was 48.8 months, whereas those concurrently underweight and wasting experienced a median of 45.4 months in this state. Additionally, rural children (HR = 1.75; 95% CI: 1.53-1.97), those with illiterate fathers (HR = 1.50; 95% CI: 1.38-1.62) and mothers (HR = 1.45; 95% CI: 1.02-3.29), and those in households lacking safe drinking water (HR = 1.70; 95% CI: 1.26-2.14) or access to cooking fuel (HR = 1.95; 95% CI: 1.75-2.17) exhibited a higher risk of undernutrition and a slower recovery rate.

Conclusions: This study revealed that rural children, especially those with illiterate parents and households lacking safe drinking water but cooking fuels, face an increased risk of undernutrition and slower recovery.

Background: A Bayesian approach may be useful in the study of possible treatment-related rare serious adverse events, particularly when there are strongly held opinions in the absence of good quality previous data. We demonstrate the application of a Bayesian analysis by integrating expert opinions with population-based epidemiologic data to investigate the association between chiropractic care and acute lumbar disc herniation (LDH) with early surgery.

Methods: Experts' opinions were used to derive probability distributions of the incidence rate ratio (IRR) for acute LDH requiring early surgery associated with chiropractic care. A 'community of priors' (enthusiastic, neutral, and skeptical) was built by dividing the experts into three groups according to their perceived mean prior IRR. The likelihood was formed from the results of a population-based epidemiologic study comparing the relative incidence of acute LDH with early surgery after chiropractic care versus primary medical care, with sensitive and specific outcome case definitions and surgery occurring within 8- and 12-week time windows after acute LDH. The robustness of results to the community of priors and specific versus sensitive case definitions was assessed.

Results: The most enthusiastic 25% of experts had a prior IRR of 0.42 (95% credible interval [CrI], 0.03 to 1.27), while the most skeptical 25% of experts had a prior IRR of 1.66 (95% CrI, 0.55 to 4.25). The Bayesian posterior estimates across priors and outcome definitions ranged from an IRR of 0.39 (95% CrI, 0.21 to 0.68) to an IRR of 1.40 (95% CrI, 0.52 to 2.55). With a sensitive definition of the outcome, the analysis produced results that confirmed prior enthusiasts' beliefs and that were precise enough to shift prior beliefs of skeptics. With a specific definition of the outcome, the results were not strong enough to overcome prior skepticism.

Conclusion: A Bayesian analysis integrating expert beliefs highlighted the value of eliciting informative priors to better understand how new evidence ought to update prior existing beliefs. Clinical epidemiologists are encouraged to integrate informative and expert opinions representing the end-user community of priors in Bayesian analyses, particularly when there are strongly held opinions in the absence of definitive scientific evidence.

Background: The CapaCiTY programme includes three, multi-centre, randomised controlled trials aiming to develop an evidence based adult chronic constipation treatment pathway. The trials were conducted in the United Kingdom, National Health Service, aiming to recruit 808 participants from 26 March 2015 to 31 January 2019. Sites were selected based on their responses to site feasibility questionnaires (2014-2015), a common tool employed by sponsors to assess a site's recruitment potential and ability to undertake the trial protocol. Failure to recruit the planned sample jeopardises reliability of results and wastes significant time and resources. The purpose of this study was to investigate barriers to recruitment in 2017.

Methods: We conducted site feasibility assessments with thirty-nine sites prior to trial commencement. Twenty-seven were selected to participate in the CapaCiTY programme, twelve were deemed unsuitable. We compared site contracted recruitment rates with actual recruitment rates and conducted a telephone survey and analysis from 5 July to 7 December 2017 (n = 24) to understand barriers to recruitment. Three sites declined to participate in the survey.

Results: At the time of survey, 15% of sites in the CapaCiTY programme were meeting recruitment targets, 85% were recruiting half or less of their target. Of these, 28% recruited no participants. The main barriers to recruitment were lack of resources, high workloads, lack of suitable participants and study design not being compatible with routine care. Despite multiple strategies employed to overcome these barriers, the trials were eventually stopped due to futility, recruiting only 34% of the programme sample size.

Conclusions: Improving the reliability of site feasibility assessments could potentially save a substantial amount in failed research investments and speed up the time to delivery of new treatments. We recommend 1) investment in training researchers in conducting and completing site feasibility; 2) funders to require pilot and feasibility data in grant applications, with an emphasis on patient and public involvement in trial design; 3) conducting site feasibility assessment at the pre-award stage; 4) development of a national database of sites' previous trial recruitment performance; 5) data-driven site level assessment of recruitment potential.

Trial registration: ISRCTN11791740; 16/07/2015, ISRCTN11093872; 11/11/2015, ISRCTN11747152; 30/09/2015.

Background: Bias from data missing not at random (MNAR) is a persistent concern in health-related research. A bias analysis quantitatively assesses how conclusions change under different assumptions about missingness using bias parameters that govern the magnitude and direction of the bias. Probabilistic bias analysis specifies a prior distribution for these parameters, explicitly incorporating available information and uncertainty about their true values. A Bayesian bias analysis combines the prior distribution with the data's likelihood function whilst a Monte Carlo bias analysis samples the bias parameters directly from the prior distribution. No study has compared a Monte Carlo bias analysis to a Bayesian bias analysis in the context of MNAR missingness.

Methods: We illustrate an accessible probabilistic bias analysis using the Monte Carlo bias analysis approach and a well-known imputation method. We designed a simulation study based on a motivating example from the UK Biobank study, where a large proportion of the outcome was missing and missingness was suspected to be MNAR. We compared the performance of our Monte Carlo bias analysis to a principled Bayesian bias analysis, complete case analysis (CCA) and multiple imputation (MI) assuming missing at random.

Results: As expected, given the simulation study design, CCA and MI estimates were substantially biased, with 95% confidence interval coverages of 7-48%. Including auxiliary variables (i.e., variables not included in the substantive analysis that are predictive of missingness and the missing data) in MI's imputation model amplified the bias due to assuming missing at random. With reasonably accurate and precise information about the bias parameter, the Monte Carlo bias analysis performed as well as the Bayesian bias analysis. However, when very limited information was provided about the bias parameter, only the Bayesian bias analysis was able to eliminate most of the bias due to MNAR whilst the Monte Carlo bias analysis performed no better than the CCA and MI.

Conclusion: The Monte Carlo bias analysis we describe is easy to implement in standard software and, in the setting we explored, is a viable alternative to a Bayesian bias analysis. We caution careful consideration of choice of auxiliary variables when applying imputation where data may be MNAR.