A Risk Calculator for Post-Operative Urinary Retention (POUR) Following Vaginal Pelvic Floor Surgery: Multivariable Prediction Modelling

Objective: To determine the peri-operative characteristics associated with an increased risk of post-operative urinary retention (POUR) following vaginal pelvic floor surgery. Design: A retrospective cohort study using multivariable prediction modelling. Setting: A tertiary referral urogynaecology unit. Population: Patients undergoing vaginal pelvic floor surgery from January 2015 to February 2020. Methods: Eighteen variables (24 parameters) were compared between those with and without POUR and then included as potential predictors in statistical models to predict POUR. The final model was chosen as the one with the largest c-index from internal cross-validation. This was then externally validated using a separate data set (n=94) from another surgical centre. Main Outcome Measures: diagnosis of POUR following surgery while the patient was in hospital. Results: Among the 700 women undergoing surgery, 301 (43%) experienced POUR. Pre-operative variables with statistically significant univariate relationships with POUR included age, menopausal status, prolapse stage, and uroflow parameters. Significant peri-operative factors included estimated blood loss, amount of intravenous fluid administered, operative time, length of stay, and specific procedures including vaginal hysterectomy with intraperitoneal vault suspension, anterior colporrhaphy, posterior colporrhaphy, and colpocleisis. The lasso logistic regression model had the best combination of internally cross-validated c-index (0.73) and accurate calibration curve. Using this data, a POUR risk calculator was developed (https://pourrisk.shinyapps.io/POUR/). Conclusions: This POUR risk calculator will allow physicians to counsel patients pre-operatively on their risk of developing POUR after vaginal pelvic surgery and help focus discussion around potential management options.

Prediction modelling in the early detection of neonatal sepsis

Background Prediction modelling can greatly assist the health-care professionals in the management of diseases, thus sparking interest in neonatal sepsis diagnosis. The main objective of the study was to provide a complete picture of performance of prediction models for early detection of neonatal sepsis. Methods PubMed, Scopus, CINAHL databases were searched and articles which used various prediction modelling measures for the early detection of neonatal sepsis were comprehended. Data extraction was carried out based on Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies checklist. Extricate data consisted of objective, study design, patient characteristics, type of statistical model, predictors, outcome, sample size and location. Prediction model Risk of Bias Assessment Tool was applied to gauge the risk of bias of the articles. Results An aggregate of ten studies were included in the review among which eight studies had applied logistic regression to build a prediction model, while the remaining two had applied artificial intelligence. Potential predictors like neonatal fever, birth weight, foetal morbidity and gender, cervicovaginitis and maternal age were identified for the early detection of neonatal sepsis. Moreover, birth weight, endotracheal intubation, thyroid hypofunction and umbilical venous catheter were promising factors for predicting late-onset sepsis; while gestational age, intrapartum temperature and antibiotics treatment were utilised as budding prognosticators for early-onset sepsis detection. Conclusion Prediction modelling approaches were able to recognise promising maternal, neonatal and laboratory predictors in the rapid detection of early and late neonatal sepsis and thus, can be considered as a novel way for clinician decision-making towards the disease diagnosis if not used alone, in the years to come.

Protocol for Development of a Reporting Guideline for Causal and Counterfactual Prediction Models

[Introduction] While there are protocols for reporting on observational studies (e.g., STROBE, RECORD), estimation of causal effects from both observational data and randomized experiments (e.g., AGREMA, CONSORT), and on prediction modelling(e.g., TRIPOD), none is purposely made for assessing the ability and reliability of models to predict counterfactuals for individuals upon one or more possible interventions, on the basis of given (or inferred) causal structures. This paper describes methods and processes that will be used to develop a reporting guideline for causal and counterfactual prediction models(tentative acronym: PRECOG). [Materials and Methods] PRECOG will be developed following published guidance from the EQUATOR network, and will comprise five stages. Stage 1 will be bi-weekly meetings of a working group with external advisors (active until stage 5). Stage 2 will comprise a scoping/systematic review of literature on counterfactual prediction modelling for biomedical sciences(registered in PROSPERO). In stage 3, we will perform a computer-based, real-time Delphi survey to consolidate the PRECOGchecklist, involving experts in causal inference, statistics, machine learning, prediction modelling and protocols/standards. Stage 4 will involve the write-up of the PRECOG guideline (including its checklist) based on the results from the prior stages. In stage 5, we will work on the publication of the guideline and of the scoping/systematic review as peer-reviewed, open-access papers, and on their dissemination through conferences, websites, and social media. [Conclusions] PRECOG can help researchers and policymakers to carry out and critically appraise causal and counterfactual prediction model studies. PRECOG will also be useful for designing interventions, and we anticipate further expansion of the guideline for specific areas, e.g., pharmaceutical interventions.

Improving service use through prediction modelling: a case study of a mathematics support centre

In higher education, student learning support centres are examples of walk-in services with nonstationary demand. For many centres, the major expenditure is tutor wages; thus, optimizing tutor numbers and ensuring value for money in this area are key. In University College Dublin, the mathematics support centre (MSC) has developed a software system, which electronically records the time each student enters the queue, their start time with a tutor and time spent with a tutor. In this paper, we show how data analysis of 25,702 student visits and tutor timetable data, spanning 6 years, is used to identify busy and quiet periods. Prediction modelling is then used to estimate the waiting time for future MSC visitors. Subsequently, we discuss how this is used for staffing optimization, i.e. to ensure there is sufficient coverage for busy times and no resource wastage during quieter periods. The analysis described resulted in the MSC reducing the number of queue abandonments and releasing funds from overstaffed hours to increase opening hours. The methods used are easily adapted for any busy walk-in service, and the code and data referenced are freely available: https://github.com/ehoward1/Math-Support-Centre-.