Dr David McLernon

Biography

I have been a Senior Research Fellow in the Biostatistics and Health Data Science group in the Institute of Applied Health Sciences for 6 years. I was a research fellow at the same institute for the previous 11 years. I have a strong methodological interest in clinical prediction modelling, particularly within the clinical area of reproductive medicine, a field I have worked in for over 15 years. One of the projects I led concerned the dynamic prediction of live birth in couples who have unexplained infertility. Such models help clinicians to counsel couples on their changing prognosis over time and may help aid decisions around when their patients should undergo ART. I led the development of the OPIS predictions models which estimate the chance of having a baby over multiple complete cycles of IVF. I also collaborate with the USA's Society for Assisted Reproductive Technology and together we developed the new SART IVF predictor at sart.org.

I am also a member of an international body called The STRATOS initiative (STRengthening Analytical Thinking for Observational Studies). Specifically, I am a member of the topic group 6 on ‘Evaluating diagnostic tests and prediction models’ (https://www.stratos-initiative.org/groups). As part of this initiative I write guidance papers for researchers on how to develop and validate (assess the predictive performance) survival prediction models.

I supervise PhD and MD students, and teach medical statistics at undergraduate and postgraduate level. 

Over the past 13 months I have taken on leadership roles within the Institute of Applied Health Sciences (IAHS). I was interim research lead from Feb to Oct 2024 and am now IAHS and Institute of Education in Healthcare and Medical Sciences REF Impact Lead

Research Overview

Methodologically, I am interested in the development of prognostic models, particularly those using advanced survival analysis methods such as accelerated failure time models and more flexible approaches. I am involved in promoting the good practice of the application of methods for developing and validating prediction models with the STRATOS Initiative Topic Group 6: Evaluating diagnostic tests and prediction models. STRATOS stands for STRengthening Analytical Thinking for Observational Studies and is an international collaboration that provides guidance to help bridge the gap between methodology and application. Prediction modelling is important because they facilitate clinical decision-making, inform patients of their probability of an outcome of interest, and can be used for risk stratification to inform RCT designs. 

Much of my work focuses on the application of statistical methods to the discipline of reproductive medicine, a field I have worked in for 15 years. One of the projects I led concerned the dynamic prediction of live birth in couples who have unexplained infertility. Such models help clinicians to counsel couples on their changing prognosis over time and may help aid decisions around when their patients should undergo ART. I led the development of the OPIS predictions models which estimate the chance of having a baby over multiple complete cycles of IVF. I also collaborated with the USA's Society for Assisted Reproductive Technology and together we developed the SART IVF predictor at sart.org.

Current Research

Predicting pregnancy outcomes in couples with infertility

Outcome Prediction In Subfertility (OPiS) models

I lead a research programme on predicting pregnancy outcomes among infertile couples, starting with a Chief Scientist Office Postdoctoral Fellowship awarded in 2012. I have developed and validated three online in-vitro fertilisation (IVF) prediction calculators for patients and clinicians in the UK. The pre-treatment calculator uses couple information from before starting IVF to inform them of their chance of success over the first complete cycle and all subsequent complete cycles up to a maximum of six. The post-treatment calculator revises this prediction when the couple have attempted their first embryo transfer. This model incorporates treatment information including the number of eggs retrieved and the number of embryos transferred. The third calculator estimates the chance of having a baby for couples about to embark on a second complete cycle of IVF. Please see: https://w3.abdn.ac.uk/clsm/opis. This tool is used by ~3000 people monthly. This work involved collaborations with clinical and methodological experts from Leiden University Medical Center and Utrecht Medical Center. Another tool was developed for the USA which was commissioned by the Society for Assisted Reproductive Therapy www.sart.org. This is used by ~8000 people per month.  

Dynamic predictions for couples with unexplained infertility

In recent previous work, we developed a novel prototype dynamic model that can predict natural and treatment related conception in couples with unexplained infertility,9 and converted it into an online tool (https://w3.abdn.ac.uk/clsm/test/tool/opis1). The model predicts from diagnosis and recalculates these chances monthly. This model was based on a relatively small dataset from Aberdeen Fertility Clinic using data from 1998-2011 (1316 couples). This work involved collaborations with clinical and methodological experts from Academic Medical Center in Amsterdam, Leiden University Medical Center and Utrecht Medical Center.

While predictions under expectant management validated well in a Dutch cohort, the precision of predictions following IVF was relatively low due to small numbers of monthly treatments. We propose to refine this model by including the following eight years’ worth of data (~2000 more couples) which will further improve the accuracy and precision of predictions. We will externally validate the improved model in a definitive cohort study using a large sample of new couples with unexplained infertility across the UK. This model will provide a pragmatic approach to personalised management for infertile couples with respect to who and when to treat.

We have also undertaken a usability study of our dynamic prediction tool. We recruited eight female public participants in order to gain a better understanding of user-interface design and interaction. The main feedback was around making the tool more user-friendly e.g. less clinical sounding text, more attractive colouring and interactive elements. The actual science behind the tool was seen as very useful and helpful, especially the personalised predictions and the fact that they are updated monthly. The outcome of this usability study helped assess design priorities and will enable us to design a better user interface for further release. 

Other projects

As well as fertility-based research, I am involved in studies from different clinical disciplines (e.g. stroke, renal disease, obstetrics and gynaecology and neurology) using a variety of statistical methodology (e.g. IPD-MA, advanced survival methods, mixed effects models). 

Past Research

In January 2012 I was awarded a Chief Scientist Office Postdoctoral Fellowship in Health Services and Health of the Public Research supervised by Prof Bhattacharya and Prof Lee.  The fellowship aims to develop clinical prediction models that can estimate the probability of pregnancy outcomes in couples attending a fertility clinic. A brief summary follows: 

Development and validation of prognostic models for subfertility

Subfertility is defined as the inability to conceive within one year of unprotected vaginal intercourse. Subfertility is a large public health burden, affecting 1 in 5 Scottish couples. Debate exists on what characteristics determine whether and when a subfertile couple should be treated. Treatment for those with a high chance of conceiving naturally may not be cost-effective and result in undue stress. Conversely, those with a low chance of spontaneous conception need immediate treatment. Whilst clinical prediction models exist that estimate the chances of treatment independent pregnancy or treatment dependent pregnancy, none take into account both outcomes.  Predicting the chance of pregnancy following IVF treatment is also important since couples with different characteristics e.g. different ages, type of infertility, time spent trying to conceive etc. will have different chances of conception. Some may require IVF sooner while others may need more cycles of IVF treatment.  The aims of this study are: 1. To develop and validate prediction models to identify who needs fertility treatment and when (Outcome Prediction In Subfertility (OPIS) Grampian); and 2. To develop clinical prediction models to predict pregnancy outcomes in women considering IVF treatment in the UK (OPIS IVF). 

For the OPIS Grampian models, world renowned databases from the Aberdeen Fertility Centre (AFC) and Aberdeen Maternity and Neonatal Databank will be utilised to identify all subfertile couples living in Grampian, who attended the Centre from 1992 to 2011. These databases contain important baseline predictors, treatment information, and outcomes (natural or treatment based pregnancy, live birth). Cox proportional hazards models will be used to estimate the chance of treatment independent pregnancy within different follow-up time periods.  A stratified medicine modelling approach will be considered to examine the absolute difference in pregnancy chances between women who undergo treatment and those who undergo expectant management. By calculating this absolute difference for women with the same baseline characteristics, the clinician can decide whether it is beneficial to treat or not.  In order to inform clinical decisions regarding when to treat, a dynamic prediction model using landmarking will be used.  This involves fitting sequential Cox models predicting treatment independent pregnancy over the following year from incremental time origins from baseline.  This enables one to determine the change in the predicted probability of pregnancy the longer treatment is delayed.  For OPIS IVF, models predicting treatment related outcomes (including preterm birth) for separate causes of subfertility following IVF will be derived using UK wide data from the Human Fertilisation and Embryology Authority.  To adjust for multiple IVF cycles a Poisson regression model will be used considering event time to be the number of cycles until success.  This removes the need to account for within-subject correlation because the woman is the unit of observation, rather than the cycle.  Predictive ability for all models will be assessed using discrimination and calibration tests.  For OPIS Grampian, external validation will be performed using an independent cohort of Dutch subfertile patients. A web-based clinical prediction aid will be created to facilitate clinical management of subfertile couples.  Such models will help clinicians and couples make individual decisions about whether fertility treatment would be beneficial and when it should be offered.

Acknowledgement

I was awarded the inaugural Backett Weir Russell Career Development Fellowship in November 2010, an internal competitive award funded by Professor Roy Weir.  This Postdoctoral Fellowship provided the opportunity to publish papers from my PhD and to further investigate my methodological interests (focussing on survival modelling techniques in particular).  The award also facilitated the development of my external Postdoctoral Fellowship application to the CSO.  Without the kind donation by Professor Weir it would have been unlikely that I would have had the protected time to apply for this funding.

My PhD involved the follow-up of patients who have had their initial liver function tests (LFTs) in primary care, and who had no clinically obvious liver disease. Prediction models using accelerated failure time methods were derived and assessed to predict risk of liver disease, liver mortality and all cause mortality at various time points. This involved large observational datasets and record-linkage procedures.

Knowledge Exchange

I have delivered a Café Med talk to the public with my clinical colleague Prof Bhattacharya called ‘Making pregnancy predictable – from the oral pill to fertility apps’. Suttie Centre Café, Foresterhill, 24th Feb 2020

I have delivered several stand-up comedy sets at Bright Club events. The most recent one was entitled ‘The danger of trousers and other scary tales from a risk-averse statistician’ at the Explorathon festival (Feb 2024).

Collaborations

I have developed collaborations with several key researchers and clinicians from The Netherlands. Prof Ewout Steyerberg is Professor of Medical Decision Making at the Department of Public Health who is world-renowned for his research in prognostic modelling and is author of the textbook ‘Clinical Prediction Models’. Professor te Velde is an Emeritus Professor from the Department of Public Health who has written multiple papers on prognostic modelling in fertility medicine. Prof Ben Mol is Professor of Obstetrics and Gynaecology at the University of Adelaide. With these eminent Professors (including Prof Bhattacharya) I have published an opinion paper on Clinical prediction models to inform individualized decision-making in subfertile couples: a stratified medicine approach.

I am involved in promoting the application of methods for assessing the performance of prediction models with the STRATOS Initiative Topic Group 6. We are mainly a large collaboration of experts in many different areas of biostatistical research e.g. statisticians, methodologists, epidemiologists, with some additional membership categories for applied researchers and clinicians engaged in research with interests and practical experience with statistical methodology.

Funding and Grants

Teaching Responsibilities

• Course Co-ordinator and lecturer in the Intermediate Statistics Course for University/external staff on survival analysis
• Course Co-ordinator and lecturer in online PGCert Understanding and Applying Regression Models (PU5559/PU5569)
• Lecturer in annual Postgraduate Basic Statistics course
• Lecturer in annual Introduction to Medical Statistics to 3rd/4th year medical students
• Co-ordinator and lecturer for the Introduction to SPSS courses for staff 
• Lecturer in the online PGCert Applied Statistics course (PU5522)
• Tutor and marker for PU5017

Non-course Teaching Responsibilities

• Lecture to clinical academic trainees and those undertaking the Scottish Clinical Research Excellence Development Scheme (SCREDS) on “Understanding clinical prediction tools: with examples from reproductive medicine”, 9th Aug 2021
• Deliver Obstetrics & gynaecology medical staff annual lecture on medical statistics – 29th Aug 2019 ‘Clinical prediction models: what are they and how can we spot a useful one?’
• Contribute to the lunchtime student consultancy clinics
• Provide statistical consultancy to staff and PhD students in the School of Medicine & Dentistry
• Marker of exams and assignments for the Applied Statistics module of the Masters in Public Health course (PU5522, PU5559/5569 and PU5017)
• Marker of theses for MSc Public Health course
• Examiner of oral protocol presentations in the Masters in Public Health course
• Invigilator for the medical statistics exam (PU5017)
• Advisor of PhD students
• Personal tutee to undergraduate students
• Mentor to colleagues

External guest lectures