Artificial intelligence to enhance clinical value across the spectrum of cardiovascular healthcare.

Artificial intelligence (AI) is increasingly being utilized in healthcare. This article provides clinicians and researchers with a step-wise foundation for high-value AI that can be applied to a variety of different data modalities. The aim is to improve the transparency and application of AI methods, with the potential to benefit patients in routine cardiovascular care. Following a clear research hypothesis, an AI-based workflow begins with data selection and pre-processing prior to analysis, with the type of data (structured, semi-structured, or unstructured) determining what type of pre-processing steps and machine-learning algorithms are required. Algorithmic and data validation should be performed to ensure the robustness of the chosen methodology, followed by an objective evaluation of performance. Seven case studies are provided to highlight the wide variety of data modalities and clinical questions that can benefit from modern AI techniques, with a focus on applying them to cardiovascular disease management. Despite the growing use of AI, further education for healthcare workers, researchers, and the public are needed to aid understanding of how AI works and to close the existing gap in knowledge. In addition, issues regarding data access, sharing, and security must be addressed to ensure full engagement by patients and the public. The application of AI within healthcare provides an opportunity for clinicians to deliver a more personalized approach to medical care by accounting for confounders, interactions, and the rising prevalence of multi-morbidity.

Using the embryo-uterus statistical model to predict pregnancy chances by using cleavage stage morphokinetics and female age: two centre-specific prediction models and mutual validation.

BACKGROUND: The predictive capability of time-lapse monitoring (TLM) selection algorithms is influenced by patient characteristics, type and quality of data included in the analysis and the used statistical methods. Previous studies excluded DET cycles of which only one embryo implanted, introducing bias into the data. Therefore, we wanted to develop a TLM prediction model that is able to predict pregnancy chances after both single- and double embryo transfer (SET and DET). METHODS: This is a retrospective study of couples (n = 1770) undergoing an in vitro fertilization cycle at the Erasmus MC, University Medical Centre Rotterdam (clinic A) or the Reinier de Graaf Hospital (clinic B). This resulted in 2058 transferred embryos with time-lapse and pregnancy outcome information. For each dataset a prediction model was established by using the Embryo-Uterus statistical model with the number of gestational sacs as the outcome variable. This process was followed by cross-validation. RESULTS: Prediction model A (based on data of clinic A) included female age, t3-t2 and t5-t4, and model B (clinic B) included female age, t2, t3-t2 and t5-t4. Internal validation showed overfitting of model A (calibration slope 0.765 and area under the curve (AUC) 0.60), and minor overfitting of model B (slope 0.915 and AUC 0.65). External validation showed that model A was capable of predicting pregnancy in the dataset of clinic B with an AUC of 0.65 (95% CI: 0.61-0.69; slope 1.223, 95% CI: 0.903-1.561). Model B was less accurate in predicting pregnancy in the dataset of clinic A (AUC 0.60, 95% CI: 0.56-0.65; slope 0.671, 95% CI: 0.422-0.939). CONCLUSION: Our study demonstrates a novel approach to the development of a TLM prediction model by applying the EU statistical model. With further development and validation in clinical practice, our prediction model approach can aid in embryo selection and decision making for SET or DET.

Predicting pregnancy chances leading to term live birth in oligo/anovulatory women diagnosed with PCOS.

RESEARCH QUESTION: Which patient features predict the time to pregnancy (TTP) leading to term live birth in infertile women diagnosed with polycystic ovary syndrome (PCOS)? DESIGN: Prospective cohort follow-up study was completed, in which initial standardized phenotyping was conducted at two Dutch university medical centres from January 2004 to January 2014. Data were linked to the Netherlands Perinatal Registry to obtain pregnancy outcomes for each participant. All women underwent treatment according to a standardized protocol, starting with ovulation induction as first-line treatment. Predictors of pregnancies (leading to term live births) during the first year after PCOS diagnosis were evaluated. RESULTS: A total of 1779 consecutive women diagnosed with PCOS between January 2004 and January 2014 were included. In the first year following screening, 659 (37%) women with PCOS attained a pregnancy leading to term birth (≥37 weeks of gestational age). A higher chance of pregnancy was associated with race, smoking, body mass index (BMI), insulin, total testosterone and sex hormone-binding globulin (SHBG) concentrations (c-statistic = 0.59). CONCLUSIONS: Predictors of an increased chance of a live birth include White race, no current smoking, lower BMI, insulin and total testosterone concentrations, and higher SHBG concentrations. This study presents a nomogram to predict the chances of achieving a pregnancy (leading to a term live birth) within 1 year of treatment.

Forskolin-induced organoid swelling is associated with long-term cystic fibrosis disease progression.

RATIONALE: Cystic fibrosis (CF) is a monogenic life-shortening disease associated with highly variable individual disease progression which is difficult to predict. Here we assessed the association of forskolin-induced swelling (FIS) of patient-derived organoids with long-term CF disease progression in multiple organs and compared FIS with the golden standard biomarker sweat chloride concentration (SCC). METHODS: We retrieved 9-year longitudinal clinical data from the Dutch CF Registry of 173 people with mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene. Individual CFTR function was defined by FIS, measured as the relative size increase of intestinal organoids after stimulation with 0.8 µM forskolin, quantified as area under the curve (AUC). We used linear mixed-effect models and multivariable logistic regression to estimate the association of FIS with long-term forced expiratory volume in 1 s % predicted (FEV1pp) decline and development of pancreatic insufficiency, CF-related liver disease and diabetes. Within these models, FIS was compared with SCC. RESULTS: FIS was strongly associated with longitudinal changes of lung function, with an estimated difference in annual FEV1pp decline of 0.32% (95% CI 0.11-0.54%; p=0.004) per 1000-point change in AUC. Moreover, increasing FIS levels were associated with lower odds of developing pancreatic insufficiency (adjusted OR 0.18, 95% CI 0.07-0.46; p<0.001), CF-related liver disease (adjusted OR 0.18, 95% CI 0.06-0.54; p=0.002) and diabetes (adjusted OR 0.34, 95% CI 0.12-0.97; p=0.044). These associations were absent for SCC. CONCLUSION: This study exemplifies the prognostic value of a patient-derived organoid-based biomarker within a clinical setting, which is especially important for people carrying rare CFTR mutations with unclear clinical consequences.

Comparison of statistical methods for the analysis of recurrent adverse events in the presence of non-proportional hazards and unobserved heterogeneity: a simulation study.

BACKGROUND: In preventive drug trials such as intermittent preventive treatment for malaria prevention during pregnancy (IPTp), where there is repeated treatment administration, recurrence of adverse events (AEs) is expected. Challenges in modelling the risk of the AEs include accounting for time-to-AE and within-patient-correlation, beyond the conventional methods. The correlation comes from two sources; (a) individual patient unobserved heterogeneity (i.e. frailty) and (b) the dependence between AEs characterised by time-dependent treatment effects. Potential AE-dependence can be modelled via time-dependent treatment effects, event-specific baseline and event-specific random effect, while heterogeneity can be modelled via subject-specific random effect. Methods that can improve the estimation of both the unobserved heterogeneity and treatment effects can be useful in understanding the evolution of risk of AEs, especially in preventive trials where time-dependent treatment effect is expected. METHODS: Using both a simulation study and the Chloroquine for Malaria in Pregnancy (NCT01443130) trial data to demonstrate the application of the models, we investigated whether the lognormal shared frailty models with restricted cubic splines and non-proportional hazards (LSF-NPH) assumption can improve estimates for both frailty variance and treatment effect compared to the conventional inverse Gaussian shared frailty model with proportional hazard (ISF-PH), in the presence of time-dependent treatment effects and unobserved patient heterogeneity. We assessed the bias, precision gain and coverage probability of 95% confidence interval of the frailty variance estimates for the models under varying known unobserved heterogeneity, sample sizes and time-dependent effects. RESULTS: The ISF-PH model provided a better coverage probability of 95% confidence interval, less bias and less precise frailty variance estimates compared to the LSF-NPH models. The LSF-NPH models yielded unbiased hazard ratio estimates at the expense of imprecision and high mean square error compared to the ISF-PH model. CONCLUSION: The choice of the shared frailty model for the recurrent AEs analysis should be driven by the study objective. Using the LSF-NPH models is appropriate if unbiased hazard ratio estimation is of primary interest in the presence of time-dependent treatment effects. However, ISF-PH model is appropriate if unbiased frailty variance estimation is of primary interest. TRIAL REGISTRATION: ClinicalTrials.gov; NCT01443130.

Developing more generalizable prediction models from pooled studies and large clustered data sets.

Prediction models often yield inaccurate predictions for new individuals. Large data sets from pooled studies or electronic healthcare records may alleviate this with an increased sample size and variability in sample characteristics. However, existing strategies for prediction model development generally do not account for heterogeneity in predictor-outcome associations between different settings and populations. This limits the generalizability of developed models (even from large, combined, clustered data sets) and necessitates local revisions. We aim to develop methodology for producing prediction models that require less tailoring to different settings and populations. We adopt internal-external cross-validation to assess and reduce heterogeneity in models' predictive performance during the development. We propose a predictor selection algorithm that optimizes the (weighted) average performance while minimizing its variability across the hold-out clusters (or studies). Predictors are added iteratively until the estimated generalizability is optimized. We illustrate this by developing a model for predicting the risk of atrial fibrillation and updating an existing one for diagnosing deep vein thrombosis, using individual participant data from 20 cohorts (N = 10 873) and 11 diagnostic studies (N = 10 014), respectively. Meta-analysis of calibration and discrimination performance in each hold-out cluster shows that trade-offs between average and heterogeneity of performance occurred. Our methodology enables the assessment of heterogeneity of prediction model performance during model development in multiple or clustered data sets, thereby informing researchers on predictor selection to improve the generalizability to different settings and populations, and reduce the need for model tailoring. Our methodology has been implemented in the R package metamisc.

Identification of differentially expressed peptides in high-throughput proteomics data.

With the advent of high-throughput proteomics, the type and amount of data pose a significant challenge to statistical approaches used to validate current quantitative analysis. Whereas many studies focus on the analysis at the protein level, the analysis of peptide-level data provides insight into changes at the sub-protein level, including splice variants, isoforms and a range of post-translational modifications. Statistical evaluation of liquid chromatography-mass spectrometry/mass spectrometry peptide-based label-free differential data is most commonly performed using a t-test or analysis of variance, often after the application of data imputation to reduce the number of missing values. In high-throughput proteomics, statistical analysis methods and imputation techniques are difficult to evaluate, given the lack of gold standard data sets. Here, we use experimental and resampled data to evaluate the performance of four statistical analysis methods and the added value of imputation, for different numbers of biological replicates. We find that three or four replicates are the minimum requirement for high-throughput data analysis and confident assignment of significant changes. Data imputation does increase sensitivity in some cases, but leads to a much higher actual false discovery rate. Additionally, we find that empirical Bayes method (limma) achieves the highest sensitivity, and we thus recommend its use for performing differential expression analysis at the peptide level.

Pharmacogenetic interactions in amyotrophic lateral sclerosis: a step closer to a cure?

Genetic mutations related to amyotrophic lateral sclerosis (ALS) act through distinct pathophysiological pathways, which may lead to varying treatment responses. Here we assess the genetic interaction between C9orf72, UNC13A, and MOBP with creatine and valproic acid treatment in two clinical trials. Genotypic data was available for 309 of the 338 participants (91.4%). The UNC13A genotype affected mortality (p = 0.012), whereas C9orf72 repeat-expansion carriers exhibited a faster rate of decline in overall (p = 0.051) and bulbar functioning (p = 0.005). A dose-response pharmacogenetic interaction was identified between creatine and the A allele of the MOBP genotype (p = 0.027), suggesting a qualitative interaction in a recessive model (HR 3.96, p = 0.015). Not taking genetic information into account may mask evidence of response to treatment or be an unrecognized source of bias. Incorporating genetic data could help investigators to identify critical treatment clues in patients with ALS.

Individualized ovarian stimulation in IVF/ICSI treatment: it is time to stop using high FSH doses in predicted low responders.

In IVF/ICSI treatment, the FSH starting dose is often increased in predicted low responders from the belief that it improves the chance of having a baby by maximizing the number of retrieved oocytes. This intervention has been evaluated in several randomized controlled trials, and despite a slight increase in the number of oocytes-on average one to two more oocytes in the high versus standard dose group-no beneficial impact on the probability of a live birth has been demonstrated (risk difference, -0.02; 95% CI, -0.11 to 0.06). Still, many clinicians and researchers maintain a highly ingrained belief in 'the more oocytes, the better'. This is mainly based on cross-sectional studies, where the positive correlation between the number of retrieved oocytes and the probability of a live birth is interpreted as a direct causal relation. If the latter would be present, indeed, maximizing the oocyte number would benefit our patients. The current paper argues that the use of high FSH doses may not actually improve the probability of a live birth for predicted low responders undergoing IVF/ICSI treatment and exemplifies the flaws of directly using cross-sectional data to guide FSH dosing in clinical practice. Also, difficulties in the de-implementation of the increased FSH dosing strategy are discussed, which include the prioritization of intermediate outcomes (such as cycle cancellations) and the potential biases in the interpretation of study findings (such as confirmation or rescue bias).

Systematic review of statistical methods for safety data in malaria chemoprevention in pregnancy trials.

BACKGROUND: Drug safety assessments in clinical trials present unique analytical challenges. Some of these include adjusting for individual follow-up time, repeated measurements of multiple outcomes and missing data among others. Furthermore, pre-specifying appropriate analysis becomes difficult as some safety endpoints are unexpected. Although existing guidelines such as CONSORT encourage thorough reporting of adverse events (AEs) in clinical trials, they provide limited details for safety data analysis. The limited guidelines may influence suboptimal analysis by failing to account for some analysis challenges above. A typical example where such challenges exist are trials of anti-malarial drugs for malaria prevention during pregnancy. Lack of proper standardized evaluation of the safety of antimalarial drugs has limited the ability to draw conclusions about safety. Therefore, a systematic review was conducted to establish the current practice in statistical analysis for preventive antimalarial drug safety in pregnancy. METHODS: The search included five databases (PubMed, Embase, Scopus, Malaria in Pregnancy Library and Cochrane Central Register of Controlled Trials) to identify original English articles reporting Phase III randomized controlled trials (RCTs) on anti-malarial drugs for malaria prevention in pregnancy published from January 2010 to July 2019. RESULTS: Eighteen trials were included in this review that collected multiple longitudinal safety outcomes including AEs. Statistical analysis and reporting of the safety outcomes in all the trials used descriptive statistics; proportions/counts (n = 18, 100%) and mean/median (n = 2, 11.1%). Results presentation included tabular (n = 16, 88.9%) and text description (n = 2, 11.1%). Univariate inferential methods were reported in most trials (n = 16, 88.9%); including Chi square/Fisher's exact test (n = 12, 66.7%), t test (n = 2, 11.1%) and Mann-Whitney/Wilcoxon test (n = 1, 5.6%). Multivariable methods, including Poisson and negative binomial were reported in few trials (n = 3, 16.7%). Assessment of a potential link between missing efficacy data and safety outcomes was not reported in any of the trials that reported efficacy missing data (n = 7, 38.9%). CONCLUSION: The review demonstrated that statistical analysis of safety data in anti-malarial drugs for malarial chemoprevention in pregnancy RCTs is inadequate. The analyses insufficiently account for multiple safety outcomes potential dependence, follow-up time and informative missing data which can compromise anti-malarial drug safety evidence development, based on the available data.

A Systematic Comparison of Designs to Study Human Fecundity.

BACKGROUND: Several epidemiologic designs allow studying fecundability, the monthly probability of pregnancy occurrence in noncontracepting couples in the general population. These designs may, to varying extents, suffer from attenuation bias and other biases. We aimed to compare the main designs: incident and prevalent cohorts, pregnancy-based, and current duration approaches. METHODS: A realistic simulation model produced individual reproductive lives of a fictitious population. We drew random population samples according to each study design, from which the cumulative probability of pregnancy was estimated. We compared the abilities of the designs to highlight the impact of an environmental factor influencing fecundability, relying on the Cox model with censoring after 12 or 6 months. RESULTS: Regarding the estimation of the cumulative probability of pregnancy, the pregnancy-based approach was the most prone to bias. When we considered a hypothetical factor associated with a hazard ratio (HR) of pregnancy of 0.7, the estimated HR was in the 0.78-0.85 range, according to designs. This attenuation bias was largest for the prevalent cohort and smallest for the current duration approach, which had the largest variance. The bias could be limited in all designs by censoring durations at 6 months. CONCLUSION: Attenuation bias in HRs cannot be ignored in fecundability studies. Focusing on the effect of exposures during the first 6 months of unprotected intercourse through censoring removes part of this bias. For risk factors that can accurately be assessed retrospectively, retrospective fecundity designs, although biased, are not much more strongly so than logistically more intensive designs entailing follow-up.

Coronary artery calcification in middle-aged women with premature ovarian insufficiency.

OBJECTIVE: Women with premature ovarian insufficiency (POI) enter menopause before age 40. Early menopause was associated with increased risk for coronary artery disease (CAD), death from cardiovascular disease and all-cause mortality. We compared the prevalence of CAD between middle-aged women on average 10 years following the initial POI diagnosis, with a population-based cohort. DESIGN: Cross-sectional case-control study. PARTICIPANTS: Women from two Dutch University Medical Centers above 45 years of age previously diagnosed with POI (n = 98) were selected and compared with age- and race-matched controls from the Multi-Ethnic Study of Atherosclerosis (MESA). MEASUREMENTS: The primary outcome was detectable coronary artery calcium (CAC) determined by coronary computed tomography (CCT). RESULTS: Women with POI had significantly higher blood pressure, cholesterol and glucose, despite lower BMI compared to controls. Similar proportions of detectable CAC (CAC score >0 Agatston Units) were observed in women with POI and controls (POI n = 16 (16%), controls n = 52 (18%), P = 0.40 and Padj  = 0.93). In women with POI separately, we were not able to identify associations between CVD risk factors and CAC. The following CVD risk factors in controls were positively associated with CAC: age, diabetes mellitus, hypertension and LDL cholesterol. HRT use was negatively associated with CAC in controls. CONCLUSIONS: The presence of CAC did not differ significantly in women with POI around 50 years of age, compared to an age- and race-matched control group. We observe no increased calcified coronary disease in POI patients, despite the presence of unfavourable cardiovascular risk factors in these women.

Cumulative live birth rates in low-prognosis women.

STUDY QUESTION: Do cumulative live birth rates (CLBRs) over multiple IVF/ICSI cycles confirm the low prognosis in women stratified according to the POSEIDON criteria? SUMMARY ANSWER: The CLBR of low-prognosis women is ~56% over 18 months of IVF/ICSI treatment and varies between the POSEIDON groups, which is primarily attributable to the impact of female age. WHAT IS KNOWN ALREADY: The POSEIDON group recently proposed a new stratification for low-prognosis women in IVF/ICSI treatment, with the aim to define more homogenous populations for clinical trials and stimulate a patient-tailored therapeutic approach. These new criteria combine qualitative and quantitative parameters to create four groups of low-prognosis women with supposedly similar biologic characteristics. STUDY DESIGN, SIZE, DURATION: This study analyzed the data of a Dutch multicenter observational cohort study including 551 low-prognosis women, aged <44 years, who initiated IVF/ICSI treatment between 2011 and 2014 and were treated with a fixed FSH dose of 150 IU/day in the first treatment cycle. PARTICIPANTS/MATERIALS, SETTING, METHODS: Low-prognosis women were categorized into one of the POSEIDON groups based on their age (younger or older than 35 years), anti-Müllerian hormone (AMH) level (above or below 0.96 ng/ml), and the ovarian response (poor or suboptimal) in their first cycle of standard stimulation. The primary outcome was the CLBR over multiple complete IVF/ICSI cycles, including all subsequent fresh and frozen-thawed embryo transfers, within 18 months of treatment. Cumulative incidence curves were obtained using an optimistic and a conservative analytic approach. MAIN RESULTS AND THE ROLE OF CHANCE: The CLBR of the low-prognosis women was on average ~56% over 18 months of IVF/ICSI treatment. Younger unexpected poor (n = 38) and suboptimal (n = 179) responders had a CLBR of ~65% and ~68%, respectively, and younger expected poor responders (n = 65) had a CLBR of ~59%. The CLBR of older unexpected poor (n = 41) and suboptimal responders (n = 102) was ~42% and ~54%, respectively, and of older expected poor responders (n = 126) ~39%. For comparison, the CLBR of younger (n = 164) and older (n = 78) normal responders with an adequate ovarian reserve was ~72% and ~58% over 18 months of treatment, respectively. No large differences were observed in the number of fresh treatment cycles between the POSEIDON groups, with an average of two fresh cycles per woman within 18 months of follow-up. LIMITATIONS, REASONS FOR CAUTION: Small numbers in some (sub)groups reduced the precision of the estimates. However, our findings provide the first relevant indication of the CLBR of low-prognosis women in the POSEIDON groups. Small FSH dose adjustments between cycles were allowed, inducing therapeutic disparity. Yet, this is in accordance with current daily practice and increases the generalizability of our findings. WIDER IMPLICATIONS OF THE FINDINGS: The CLBRs vary between the POSEIDON groups. This heterogeneity is primarily determined by a woman's age, reflecting the importance of oocyte quality. In younger women, current IVF/ICSI treatment reaches relatively high CLBR over multiple complete cycles, despite reduced quantitative parameters. In older women, the CLBR remains relatively low over multiple complete cycles, due to the co-occurring decline in quantitative and qualitative parameters. As no effective interventions exist to counteract this decline, clinical management currently relies on proper counselling. STUDY FUNDING/COMPETING INTEREST(S): No external funds were obtained for this study. J.A.L. is supported by a Research Fellowship grant and received an unrestricted personal grant from Merck BV. S.C.O., T.C.v.T., and H.L.T. received an unrestricted personal grant from Merck BV. C.B.L. received research grants from Merck, Ferring, and Guerbet. K.F. received unrestricted research grants from Merck Serono, Ferring, and GoodLife. She also received fees for lectures and consultancy from Ferring and GoodLife. A.H. declares that the Department of Obstetrics and Gynaecology, University Medical Centre Groningen received an unrestricted research grant from Ferring Pharmaceuticals BV, the Netherlands. J.S.E.L. has received unrestricted research grants from Ferring, Zon-MW, and The Dutch Heart Association. He also received travel grants and consultancy fees from Danone, Euroscreen, Ferring, AnshLabs, and Titus Healthcare. B.W.J.M. is supported by an National Health and Medical Research Council Practitioner Fellowship (GNT1082548) and reports consultancy work for ObsEva, Merck, and Guerbet. He also received a research grant from Merck BV and travel support from Guerbet. F.J.M.B. received monetary compensation as a member of the external advisory board for Merck Serono (the Netherlands) and Ferring Pharmaceuticals BV (the Netherlands) for advisory work for Gedeon Richter (Belgium) and Roche Diagnostics on automated AMH assay development, and for a research cooperation with Ansh Labs (USA). All other authors have nothing to declare. TRIAL REGISTRATION NUMBER: Not applicable.

Critical design considerations for time-to-event endpoints in amyotrophic lateral sclerosis clinical trials.

BACKGROUND: Funding and resources for low prevalent neurodegenerative disorders such as amyotrophic lateral sclerosis (ALS) are limited, and optimising their use is vital for efficient drug development. In this study, we review the design assumptions for pivotal ALS clinical trials with time-to-event endpoints and provide optimised settings for future trials. METHODS: We extracted design settings from 13 completed placebo-controlled trials. Optimal assumptions were estimated using parametric survival models in individual participant data (n=4991). Designs were compared in terms of sample size, trial duration, drug use and costs. RESULTS: Previous trials overestimated the hazard rate by 18.9% (95% CI 3.4% to 34.5%, p=0.021). The median expected HR was 0.56 (range 0.33-0.66). Additionally, we found evidence for an increasing mean hazard rate over time (Weibull shape parameter of 2.03, 95% CI 1.93 to 2.15, p<0.001), which affects the design and planning of future clinical trials. Incorporating accrual time and assuming an increasing hazard rate at the design stage reduced sample size by 33.2% (95% CI 27.9 to 39.4), trial duration by 17.4% (95% CI 11.6 to 23.3), drug use by 14.3% (95% CI 9.6 to 19.0) and follow-up costs by 21.2% (95% CI 15.6 to 26.8). CONCLUSIONS: Implementing distributional knowledge and incorporating accrual at the design stage could achieve large gains in the efficiency of ALS clinical trials with time-to-event endpoints. We provide an open-source platform that helps investigators to make more accurate sample size calculations and optimise the use of their available resources.

Sample size considerations and predictive performance of multinomial logistic prediction models.

Multinomial Logistic Regression (MLR) has been advocated for developing clinical prediction models that distinguish between three or more unordered outcomes. We present a full-factorial simulation study to examine the predictive performance of MLR models in relation to the relative size of outcome categories, number of predictors and the number of events per variable. It is shown that MLR estimated by Maximum Likelihood yields overfitted prediction models in small to medium sized data. In most cases, the calibration and overall predictive performance of the multinomial prediction model is improved by using penalized MLR. Our simulation study also highlights the importance of events per variable in the multinomial context as well as the total sample size. As expected, our study demonstrates the need for optimism correction of the predictive performance measures when developing the multinomial logistic prediction model. We recommend the use of penalized MLR when prediction models are developed in small data sets or in medium sized data sets with a small total sample size (ie, when the sizes of the outcome categories are balanced). Finally, we present a case study in which we illustrate the development and validation of penalized and unpenalized multinomial prediction models for predicting malignancy of ovarian cancer.

Do female age and body weight modify the effect of individualized FSH dosing in IVF/ICSI treatment? A secondary analysis of the OPTIMIST trial.

INTRODUCTION: The OPTIMIST trial revealed that for women starting in vitro fertilization (IVF)/intracytoplasmic sperm injection (ICSI) treatment, no substantial differences exist in first cycle and cumulative live birth rates between an antral follicle count (AFC)-based individualized follicle-stimulating hormone (FSH) dose and a standard dose. Female age and body weight have been suggested to cause heterogeneity in the effect of FSH dose individualization. The objective of the current study is to evaluate whether these patient characteristics modify the effect of AFC-based individualized FSH dosing in IVF/ICSI treatment. MATERIAL AND METHODS: A secondary data-analysis of the OPTIMIST trial. Women initiating IVF/ICSI treatment were classified as predicted poor (AFC 0-7), suboptimal (AFC 8-10) or hyper responders (AFC >15), and randomly allocated to a standard FSH dose (150 IU/d) or an individualized FSH dose (450, 225 or 100 IU/d for predicted poor, suboptimal and hyper responders, respectively). In each predicted response category, logistic regression models with interaction terms were used to evaluate the presence of effect modification. The first cycle was analyzed, and the primary outcomes were first complete cycle live birth rate (including fresh plus frozen-thawed embryo transfers) and ovarian hyperstimulation syndrome (OHSS) risks. RESULTS: No effect modification was revealed in the predicted poor (n = 234) and suboptimal (n = 277) responders. In the predicted hyper responders (n = 521), the effect of the individualized FSH dose on the first cycle live birth rate was modified by female age (P = 0.02) and the effect on OHSS risks was modified by body weight (P = 0.02). A dose reduction from 150 to 100 IU/d generally decreased the OHSS risks in predicted hyper responders, but also reduced the chance of a live birth in young women, and had no beneficial impact on OHSS risks in women with a relatively low body weight. CONCLUSIONS: In women with a predicted hyper response undergoing IVF/ICSI treatment, female age and body weight seem to modify the effect of FSH dose individualization. Although a reduced FSH starting dose generally decreases the OHSS risks, it may also reduce the chance of a live birth, specifically for young women. Future studies could consider these findings when investigating the optimal approach to reduce OHSS risks while maintaining the probability of a live birth for predicted hyper responders in IVF/ICSI treatment.

Anti-Müllerian Hormone Trajectories Are Associated With Cardiovascular Disease in Women: Results From the Doetinchem Cohort Study.

BACKGROUND: Earlier age at menopause is widely considered to be associated with an increased risk of cardiovascular disease. However, the underlying mechanisms of this relationship remain undetermined. Indications suggest that anti-Müllerian hormone (AMH), an ovarian reserve marker, plays a physiological role outside of the reproductive system. Therefore, we investigated whether longitudinal AMH decline trajectories are associated with an increased risk of cardiovascular disease (CVD) occurrence. METHODS: This study included 3108 female participants between 20 and 60 years of age at baseline of the population-based Doetinchem Cohort. Participants completed ≥1 of 5 consecutive quinquennial visits between 1987 and 2010, resulting in a total follow-up time of 20 years. AMH was measured in 8507 stored plasma samples. Information on total CVD, stroke, and coronary heart disease was obtained through a hospital discharge registry linkage. The association of AMH trajectories with CVD was quantified with joint modeling, with adjustment for age, smoking, oral contraceptive use, body mass index, menopausal status, postmenopausal hormone therapy use, diastolic blood pressure, total cholesterol, high-density lipoprotein cholesterol, and glucose levels. RESULTS: By the end of follow-up, 8.2% of the women had suffered from CVD, 4.9% had suffered from coronary heart disease, and 2.6% had experienced a stroke. After adjustment, each ng/mL lower logAMH level was associated with a 21% higher risk of CVD (hazard ratio, 1.21; 95% confidence interval, 1.07-1.36) and a 26% higher risk of coronary heart disease (hazard ratio, 1.25; 95% confidence interval, 1.08-1.46). Each additional ng/mL/year decrease of logAMH was associated with a significantly higher risk of CVD (hazard ratio, 1.46; 95% confidence interval, 1.14-1.87) and coronary heart disease (hazard ratio, 1.56; 95% confidence interval, 1.15-2.12). No association between AMH and stroke was found. CONCLUSIONS: These results indicate that AMH trajectories in women are independently associated with CVD risk. Therefore, we postulate that the decline of circulating AMH levels may be part of the pathophysiology of the increased cardiovascular risk of earlier menopause. Confirmation of this association and elucidation of its underlying mechanisms are needed to place these results in a clinical perspective.

Monitoring disease progression with plasma creatinine in amyotrophic lateral sclerosis clinical trials.

OBJECTIVES: Plasma creatinine is a predictor of survival in amyotrophic lateral sclerosis (ALS). It remains, however, to be established whether it can monitor disease progression and serve as surrogate endpoint in clinical trials. METHODS: We used clinical trial data from three cohorts of clinical trial participants in the LITRA, EMPOWER and PROACT studies. Longitudinal associations between functional decline, muscle strength and survival with plasma creatinine were assessed. Results were translated to trial design in terms of sample size and power. RESULTS: A total of 13 564 measurements were obtained for 1241 patients. The variability between patients in rate of decline was lower in plasma creatinine than in ALS functional rating scale-Revised (ALSFRS-R; p<0.001). The average rate of decline was faster in the ALSFRS-R, with less between-patient variability at baseline (p<0.001). Plasma creatinine had strong longitudinal correlations with the ALSFRS-R (0.43 (0.39-0.46), p<0.001), muscle strength (0.55 (0.51-0.58), p<0.001) and overall mortality (HR 0.88 (0.86-0.91, p<0.001)). Using plasma creatinine as outcome could reduce the sample size in trials by 21.5% at 18 months. For trials up to 10 months, the ALSFRS-R required a lower sample size. CONCLUSIONS: Plasma creatinine is an inexpensive and easily accessible biomarker that exhibits less variability between patients with ALS over time and is predictive for the patient's functional status, muscle strength and mortality risk. Plasma creatinine may, therefore, increase the power to detect treatment effects and could be incorporated in future ALS clinical trials as potential surrogate outcome.

Maturation of the Infant Respiratory Microbiota, Environmental Drivers, and Health Consequences. A Prospective Cohort Study.

RATIONALE: Perinatal and postnatal influences are presumed important drivers of the early-life respiratory microbiota composition. We hypothesized that the respiratory microbiota composition and development in infancy is affecting microbiota stability and thereby resistance against respiratory tract infections (RTIs) over time. OBJECTIVES: To investigate common environmental drivers, including birth mode, feeding type, antibiotic exposure, and crowding conditions, in relation to respiratory tract microbiota maturation and stability, and consecutive risk of RTIs over the first year of life. METHODS: In a prospectively followed cohort of 112 infants, we characterized the nasopharyngeal microbiota longitudinally from birth on (11 consecutive sample moments and the maximum three RTI samples per subject; in total, n = 1,121 samples) by 16S-rRNA gene amplicon sequencing. MEASUREMENTS AND MAIN RESULTS: Using a microbiota-based machine-learning algorithm, we found that children experiencing a higher number of RTIs in the first year of life already demonstrate an aberrant microbial developmental trajectory from the first month of life on as compared with the reference group (0-2 RTIs/yr). The altered microbiota maturation process coincided with decreased microbial community stability, prolonged reduction of Corynebacterium and Dolosigranulum, enrichment of Moraxella very early in life, followed by later enrichment of Neisseria and Prevotella spp. Independent drivers of these aberrant developmental trajectories of respiratory microbiota members were mode of delivery, infant feeding, crowding, and recent antibiotic use. CONCLUSIONS: Our results suggest that environmental drivers impact microbiota development and, consequently, resistance against development of RTIs. This supports the idea that microbiota form the mediator between early-life environmental risk factors for and susceptibility to RTIs over the first year of life.

Meta-analysis approach as a gene selection method in class prediction: does it improve model performance? A case study in acute myeloid leukemia.

BACKGROUND: Aggregating gene expression data across experiments via meta-analysis is expected to increase the precision of the effect estimates and to increase the statistical power to detect a certain fold change. This study evaluates the potential benefit of using a meta-analysis approach as a gene selection method prior to predictive modeling in gene expression data. RESULTS: Six raw datasets from different gene expression experiments in acute myeloid leukemia (AML) and 11 different classification methods were used to build classification models to classify samples as either AML or healthy control. First, the classification models were trained on gene expression data from single experiments using conventional supervised variable selection and externally validated with the other five gene expression datasets (referred to as the individual-classification approach). Next, gene selection was performed through meta-analysis on four datasets, and predictive models were trained with the selected genes on the fifth dataset and validated on the sixth dataset. For some datasets, gene selection through meta-analysis helped classification models to achieve higher performance as compared to predictive modeling based on a single dataset; but for others, there was no major improvement. Synthetic datasets were generated from nine simulation scenarios. The effect of sample size, fold change and pairwise correlation between differentially expressed (DE) genes on the difference between MA- and individual-classification model was evaluated. The fold change and pairwise correlation significantly contributed to the difference in performance between the two methods. The gene selection via meta-analysis approach was more effective when it was conducted using a set of data with low fold change and high pairwise correlation on the DE genes. CONCLUSION: Gene selection through meta-analysis on previously published studies potentially improves the performance of a predictive model on a given gene expression data.