Mixed effect estimation in deep compartment models: Variational methods outperform first-order approximations.

This work focusses on extending the deep compartment model (DCM) framework to the estimation of mixed-effects. By introducing random effects, model predictions can be personalized based on drug measurements, enabling the testing of different treatment schedules on an individual basis. The performance of classical first-order (FO and FOCE) and machine learning based variational inference (VI) algorithms were compared in a simulation study. In VI, posterior distributions of the random variables are approximated using variational distributions whose parameters can be directly optimized. We found that variational approximations estimated using the path derivative gradient estimator version of VI were highly accurate. Models fit on the simulated data set using the FO and VI objective functions gave similar results, with accurate predictions of both the population parameters and covariate effects. Contrastingly, models fit using FOCE depicted erratic behaviour during optimization, and resulting parameter estimates were inaccurate. Finally, we compared the performance of the methods on two real-world data sets of haemophilia A patients who received standard half-life factor VIII concentrates during prophylactic and perioperative settings. Again, models fit using FO and VI depicted similar results, although some models fit using FO presented divergent results. Again, models fit using FOCE were unstable. In conclusion, we show that mixed-effects estimation using the DCM is feasible. VI performs conditional estimation, which might lead to more accurate results in more complex models compared to the FO method.

On inductive biases for the robust and interpretable prediction of drug concentrations using deep compartment models.

Conventional pharmacokinetic (PK) models contain several useful inductive biases guiding model convergence to more realistic predictions of drug concentrations. Implementing similar biases in standard neural networks can be challenging, but might be fundamental for model robustness and predictive performance. In this study, we build on the deep compartment model (DCM) architecture by introducing constraints that guide the model to explore more physiologically realistic solutions. Using a simulation study, we show that constraints improve robustness in sparse data settings. Additionally, predicted concentration-time curves took on more realistic shapes compared to unconstrained models. Next, we propose the use of multi-branch networks, where each covariate can be connected to specific PK parameters, to reduce the propensity of models to learn spurious effects. Another benefit of this architecture is that covariate effects are isolated, enabling model interpretability through the visualization of learned functions. We show that all models were sensitive to learning false effects when trained in the presence of unimportant covariates, indicating the importance of selecting an appropriate set of covariates to link to the PK parameters. Finally, we compared the predictive performance of the constrained models to previous relevant population PK models on a real-world data set of 69 haemophilia A patients. Here, constrained models obtained higher accuracy compared to the standard DCM, with the multi-branch network outperforming previous PK models. We conclude that physiological-based constraints can improve model robustness. We describe an interpretable architecture which aids model trust, which will be key for the adoption of machine learning-based models in clinical practice.

A Generative and Causal Pharmacokinetic Model for Factor VIII in Hemophilia A: A Machine Learning Framework for Continuous Model Refinement.

In rare diseases, such as hemophilia A, the development of accurate population pharmacokinetic (PK) models is often hindered by the limited availability of data. Most PK models are specific to a single recombinant factor VIII (rFVIII) concentrate or measurement assay, and are generally unsuited for answering counterfactual ("what-if") queries. Ideally, data from multiple hemophilia treatment centers are combined but this is generally difficult as patient data are kept private. In this work, we utilize causal inference techniques to produce a hybrid machine learning (ML) PK model that corrects for differences between rFVIII concentrates and measurement assays. Next, we augment this model with a generative model that can simulate realistic virtual patients as well as impute missing data. This model can be shared instead of actual patient data, resolving privacy issues. The hybrid ML-PK model was trained on chromogenic assay data of lonoctocog alfa and predictive performance was then evaluated on an external data set of patients who received octocog alfa with FVIII levels measured using the one-stage assay. The model presented higher accuracy compared with three previous PK models developed on data similar to the external data set (root mean squared error = 14.6 IU/dL vs. mean of 17.7 IU/dL). Finally, we show that the generative model can be used to accurately impute missing data (< 18% error). In conclusion, the proposed approach introduces interesting new possibilities for model development. In the context of rare disease, the introduction of generative models facilitates sharing of synthetic data, enabling the iterative improvement of population PK models.

The added value of temporal data and the best way to handle it: A use-case for atrial fibrillation using general practitioner data.

INTRODUCTION: Temporal data has numerous challenges for deep learning such as irregularity of sampling. New algorithms are being developed that can handle these temporal challenges better. However, it is unclear how the performance ranges from classical non-temporal models to newly developed algorithms. Therefore, this study compares different non-temporal and temporal algorithms for a relevant use case, the prediction of atrial fibrillation (AF) using general practitioner (GP) data. METHODS: Three datasets with a 365-day observation window and prediction windows of 14, 180 and 360 days were used. Data consisted of medication, lab, symptom, and chronic diseases codings registered by the GP. The benchmark discarded temporality and used logistic regression, XGBoost models and neural networks on the presence of codings over the whole year. Pattern data extracted common patterns of GP codings and tested using the same algorithms. LSTM and CKConv models were trained as models incorporating temporality. RESULTS: Algorithms which incorporated temporality (LSTM and CKConv, (max AUC 0.734 at 360 days prediction window) outperformed both benchmark and pattern algorithms (max AUC 0.723, with a significant improvement using the 360 days prediction window (p = 0.04). The difference between the benchmark and the LSTM or CKConv algorithm decreased with smaller prediction windows, indicating temporal importance for longer prediction windows. The CKConv and LSTM algorithm performed similarly, possibly due to limited sequence length. CONCLUSION: Temporal models outperformed non-temporal models for the prediction of AF. For temporal models, CKConv is a promising algorithm to handle temporal data using GP data as it can handle irregular data.

Early identification of persistent somatic symptoms in primary care: data-driven and theory-driven predictive modelling based on electronic medical records of Dutch general practices.

OBJECTIVE: The present study aimed to early identify patients with persistent somatic symptoms (PSS) in primary care by exploring routine care data-based approaches. DESIGN/SETTING: A cohort study based on routine primary care data from 76 general practices in the Netherlands was executed for predictive modelling. PARTICIPANTS: Inclusion of 94 440 adult patients was based on: at least 7-year general practice enrolment, having more than one symptom/disease registration and >10 consultations. METHODS: Cases were selected based on the first PSS registration in 2017-2018. Candidate predictors were selected 2-5 years prior to PSS and categorised into data-driven approaches: symptoms/diseases, medications, referrals, sequential patterns and changing lab results; and theory-driven approaches: constructed factors based on literature and terminology in free text. Of these, 12 candidate predictor categories were formed and used to develop prediction models by cross-validated least absolute shrinkage and selection operator regression on 80% of the dataset. Derived models were internally validated on the remaining 20% of the dataset. RESULTS: All models had comparable predictive values (area under the receiver operating characteristic curves=0.70 to 0.72). Predictors are related to genital complaints, specific symptoms (eg, digestive, fatigue and mood), healthcare utilisation, and number of complaints. Most fruitful predictor categories are literature-based and medications. Predictors often had overlapping constructs, such as digestive symptoms (symptom/disease codes) and drugs for anti-constipation (medication codes), indicating that registration is inconsistent between general practitioners (GPs). CONCLUSIONS: The findings indicate low to moderate diagnostic accuracy for early identification of PSS based on routine primary care data. Nonetheless, simple clinical decision rules based on structured symptom/disease or medication codes could possibly be an efficient way to support GPs in identifying patients at risk of PSS. A full data-based prediction currently appears to be hampered by inconsistent and missing registrations. Future research on predictive modelling of PSS using routine care data should focus on data enrichment or free-text mining to overcome inconsistent registrations and improve predictive accuracy.

Embracing cohort heterogeneity in clinical machine learning development: a step toward generalizable models.

This study is a simple illustration of the benefit of averaging over cohorts, rather than developing a prediction model from a single cohort. We show that models trained on data from multiple cohorts can perform significantly better in new settings than models based on the same amount of training data but from just a single cohort. Although this concept seems simple and obvious, no current prediction model development guidelines recommend such an approach.

Evolution of Clinical Phenotypes of COVID-19 Patients During Intensive Care Treatment: An Unsupervised Machine Learning Analysis.

BACKGROUND: Identification of clinical phenotypes in critically ill COVID-19 patients could improve understanding of the disease heterogeneity and enable prognostic and predictive enrichment. However, previous attempts did not take into account temporal dynamics with high granularity. By including the dimension of time, we aim to gain further insights into the heterogeneity of COVID-19. METHODS: We used granular data from 3202 adult COVID patients in the Dutch Data Warehouse that were admitted to one of 25 Dutch ICUs between February 2020 and March 2021. Parameters including demographics, clinical observations, medications, laboratory values, vital signs, and data from life support devices were selected. Twenty-one datasets were created that each covered 24 h of ICU data for each day of ICU treatment. Clinical phenotypes in each dataset were identified by performing cluster analyses. Both evolution of the clinical phenotypes over time and patient allocation to these clusters over time were tracked. RESULTS: The final patient cohort consisted of 2438 COVID-19 patients with a ICU mortality outcome. Forty-one parameters were chosen for cluster analysis. On admission, both a mild and a severe clinical phenotype were found. After day 4, the severe phenotype split into an intermediate and a severe phenotype for 11 consecutive days. Heterogeneity between phenotypes appears to be driven by inflammation and dead space ventilation. During the 21-day period, only 8.2% and 4.6% of patients in the initial mild and severe clusters remained assigned to the same phenotype respectively. The clinical phenotype half-life was between 5 and 6 days for the mild and severe phenotypes, and about 3 days for the medium severe phenotype. CONCLUSIONS: Patients typically do not remain in the same cluster throughout intensive care treatment. This may have important implications for prognostic or predictive enrichment. Prominent dissimilarities between clinical phenotypes are predominantly driven by inflammation and dead space ventilation.

Adoption of Machine Learning in Pharmacometrics: An Overview of Recent Implementations and Their Considerations.

Pharmacometrics is a multidisciplinary field utilizing mathematical models of physiology, pharmacology, and disease to describe and quantify the interactions between medication and patient. As these models become more and more advanced, the need for advanced data analysis tools grows. Recently, there has been much interest in the adoption of machine learning (ML) algorithms. These algorithms offer strong function approximation capabilities and might reduce the time spent on model development. However, ML tools are not yet an integral part of the pharmacometrics workflow. The goal of this work is to discuss how ML algorithms have been applied in four stages of the pharmacometrics pipeline: data preparation, hypothesis generation, predictive modelling, and model validation. We will also discuss considerations before the use of ML algorithms with respect to each topic. We conclude by summarizing applications that hold potential for adoption by pharmacometricians.

Prediction of heart failure 1 year before diagnosis in general practitioner patients using machine learning algorithms: a retrospective case-control study.

OBJECTIVES: Heart failure (HF) is a commonly occurring health problem with high mortality and morbidity. If potential cases could be detected earlier, it may be possible to intervene earlier, which may slow progression in some patients. Preferably, it is desired to reuse already measured data for screening of all persons in an age group, such as general practitioner (GP) data. Furthermore, it is essential to evaluate the number of people needed to screen to find one patient using true incidence rates, as this indicates the generalisability in the true population. Therefore, we aim to create a machine learning model for the prediction of HF using GP data and evaluate the number needed to screen with true incidence rates. DESIGN, SETTINGS AND PARTICIPANTS: GP data from 8543 patients (-2 to -1 year before diagnosis) and controls aged 70+ years were obtained retrospectively from 01 January 2012 to 31 December 2019 from the Nivel Primary Care Database. Codes about chronic illness, complaints, diagnostics and medication were obtained. Data were split in a train/test set. Datasets describing demographics, the presence of codes (non-sequential) and upon each other following codes (sequential) were created. Logistic regression, random forest and XGBoost models were trained. Predicted outcome was the presence of HF after 1 year. The ratio case:control in the test set matched true incidence rates (1:45). RESULTS: Sole demographics performed average (area under the curve (AUC) 0.692, CI 0.677 to 0.706). Adding non-sequential information combined with a logistic regression model performed best and significantly improved performance (AUC 0.772, CI 0.759 to 0.785, p<0.001). Further adding sequential information did not alter performance significantly (AUC 0.767, CI 0.754 to 0.780, p=0.07). The number needed to screen dropped from 14.11 to 5.99 false positives per true positive. CONCLUSION: This study created a model able to identify patients with pending HF a year before diagnosis.

Pulmonary pathophysiology development of COVID-19 assessed by serial Electrical Impedance Tomography in the MaastrICCht cohort.

Patients with SARS-CoV-2 infection present with different lung compliance and progression of disease differs. Measures of lung mechanics in SARS-CoV-2 patients may unravel different pathophysiologic mechanisms during mechanical ventilation. The objective of this prospective observational study is to describe whether Electrical Impedance Tomography (EIT) guided positive end-expiratory pressure (PEEP) levels unravel changes in EIT-derived parameters over time and whether the changes differ between survivors and non-survivors. Serial EIT-measurements of alveolar overdistension, collapse, and compliance change in ventilated SARS-CoV-2 patients were analysed. In 80 out of 94 patients, we took 283 EIT measurements (93 from day 1-3 after intubation, 66 from day 4-6, and 124 from day 7 and beyond). Fifty-one patients (64%) survived the ICU. At admission mean PaO2/FiO2-ratio was 184.3 (SD 61.4) vs. 151.3 (SD 54.4) mmHg, (p = 0.017) and PEEP was 11.8 (SD 2.8) cmH2O vs. 11.3 (SD 3.4) cmH2O, (p = 0.475), for ICU survivors and non-survivors. At day 1-3, compliance was ~ 55 mL/cmH2O vs. ~ 45 mL/cmH2O in survivors vs. non-survivors. The intersection of overdistension and collapse curves appeared similar at a PEEP of ~ 12-13 cmH2O. At day 4-6 compliance changed to ~ 50 mL/cmH2O vs. ~ 38 mL/cmH2O. At day 7 and beyond, compliance was ~ 38 mL/cmH2O with the intersection at a PEEP of ~ 9 cmH2O vs. ~ 25 mL/cmH2O with overdistension intersecting at collapse curves at a PEEP of ~ 7 cmH2O. Surviving SARS-CoV-2 patients show more favourable EIT-derived parameters and a higher compliance compared to non-survivors over time. This knowledge is valuable for discovering the different groups.

Detection of primary Sjögren's syndrome in primary care: developing a classification model with the use of routine healthcare data and machine learning.

BACKGROUND: Primary Sjögren's Syndrome (pSS) is a rare autoimmune disease that is difficult to diagnose due to a variety of clinical presentations, resulting in misdiagnosis and late referral to specialists. To improve early-stage disease recognition, this study aimed to develop an algorithm to identify possible pSS patients in primary care. We built a machine learning algorithm which was based on combined healthcare data as a first step towards a clinical decision support system. METHOD: Routine healthcare data, consisting of primary care electronic health records (EHRs) data and hospital claims data (HCD), were linked on patient level and consisted of 1411 pSS and 929,179 non-pSS patients. Logistic regression (LR) and random forest (RF) models were used to classify patients using age, gender, diseases and symptoms, prescriptions and GP visits. RESULTS: The LR and RF models had an AUC of 0.82 and 0.84, respectively. Many actual pSS patients were found (sensitivity LR = 72.3%, RF = 70.1%), specificity was 74.0% (LR) and 77.9% (RF) and the negative predictive value was 99.9% for both models. However, most patients classified as pSS patients did not have a diagnosis of pSS in secondary care (positive predictive value LR = 0.4%, RF = 0.5%). CONCLUSION: This is the first study to use machine learning to classify patients with pSS in primary care using GP EHR data. Our algorithm has the potential to support the early recognition of pSS in primary care and should be validated and optimized in clinical practice. To further enhance the algorithm in detecting pSS in primary care, we suggest it is improved by working with experienced clinicians.

Diagnostic stewardship for blood cultures in the emergency department: A multicenter validation and prospective evaluation of a machine learning prediction tool.

BACKGROUND: Overuse of blood cultures (BCs) in emergency departments (EDs) leads to low yields and high numbers of contaminated cultures, accompanied by increased diagnostics, antibiotic usage, prolonged hospitalization, and mortality. We aimed to simplify and validate a recently developed machine learning model to help safely withhold BC testing in low-risk patients. METHODS: We extracted data from the electronic health records (EHR) for 44.123 unique ED visits with BC sampling in the Amsterdam UMC (locations VUMC and AMC; the Netherlands), Zaans Medical Center (ZMC; the Netherlands), and Beth Israel Deaconess Medical Center (BIDMC; United States) in periods between 2011 and 2021. We trained a machine learning model on the VUMC data to predict blood culture outcomes and validated it in the AMC, ZMC, and BIDMC with subsequent real-time prospective evaluation in the VUMC. FINDINGS: The model had an Area Under the Receiver Operating Characteristics curve (AUROC) of 0.81 (95%-CI = 0.78-0.83) in the VUMC test set. The most important predictors were temperature, creatinine, and C-reactive protein. The AUROCs in the validation cohorts were 0.80 (AMC; 0.78-0.82), 0.76 (ZMC; 0.74-0.78), and 0.75 (BIDMC; 0.74-0.76). During real-time prospective evaluation in the EHR of the VUMC, it reached an AUROC of 0.76 (0.71-0.81) among 590 patients with BC draws in the ED. The prospective evaluation showed that the model can be used to safely withhold blood culture analyses in at least 30% of patients in the ED. INTERPRETATION: We developed a machine learning model to predict blood culture outcomes in the ED, which retained its performance during external validation and real-time prospective evaluation. Our model can identify patients at low risk of having a positive blood culture. Using the model in practice can significantly reduce the number of blood culture analyses and thus avoid the hidden costs of false-positive culture results. FUNDING: This research project was funded by the Amsterdam Public Health - Quality of Care program and the Dutch "Doen of Laten" project (project number: 839205002).

Ratio of arterial blood pressures at borders of window surrounding systolic peak indicates patent ductus arteriosus in preterm infants.

OBJECTIVE: Presence of a patent ductus arteriosus (PDA) in neonates is assessed by echocardiography. Echocardiographic assessment has disadvantages, primarily its discontinuous nature. We hypothesize that the continuously measured ratio of arterial blood pressures (ABP) at the borders of a window surrounding the systolic peak ratio discriminates non-PDA from PDA patients. APPROACH: Preterm infants (gestational age <32 weeks) with and without PDA were included. Patients were divided into controls (n = 8) and PDA patients (n = 22), the latter with a subset of patients with closed PDA after three doses Ibuprofen (n = 10). For each patient, a six-hour ABP segment from 12 AM to 6 AM on the day of echocardiographic assessment patency or closure of the DA was selected. The mean ratio of the ABP values a samples before and p samples after the systolic peak (R ABP) was calculated for each segment. If R ABP < 1, the patient was predicted to have a PDA. The a and p with the least misclassifications were selected (-64 and +104 ms). MAIN RESULTS: R ABP was significantly lower in PDA patients (median 0.95, IQR 0.06) compared to controls (median 1.05, IQR 0.10; p = 0.0024). R ABP correctly predicted 19 out of 22 patients (86.4%) and six out of eight controls (75%). R ABP increased after closure in nine out of 10 patients (median 1.01, IQR 0.04; p = 0. 0182). SIGNIFICANCE: R ABP may discriminate preterm PDA patients from non-PDA patients and can be calculated continuously from clinical data measured during standard of care.

Serial measurements in COVID-19-induced acute respiratory disease to unravel heterogeneity of the disease course: design of the Maastricht Intensive Care COVID cohort (MaastrICCht).

INTRODUCTION: The course of the disease in SARS-CoV-2 infection in mechanically ventilated patients is unknown. To unravel the clinical heterogeneity of the SARS-CoV-2 infection in these patients, we designed the prospective observational Maastricht Intensive Care COVID cohort (MaastrICCht). We incorporated serial measurements that harbour aetiological, diagnostic and predictive information. The study aims to investigate the heterogeneity of the natural course of critically ill patients with a SARS-CoV-2 infection. METHODS AND ANALYSIS: Mechanically ventilated patients admitted to the intensive care with a SARS-CoV-2 infection will be included. We will collect clinical variables, vital parameters, laboratory variables, mechanical ventilator settings, chest electrical impedance tomography, ECGs, echocardiography as well as other imaging modalities to assess heterogeneity of the course of a SARS-CoV-2 infection in critically ill patients. The MaastrICCht is also designed to foster various other studies and registries and intends to create an open-source database for investigators. Therefore, a major part of the data collection is aligned with an existing national intensive care data registry and two international COVID-19 data collection initiatives. Additionally, we create a flexible design, so that additional measures can be added during the ongoing study based on new knowledge obtained from the rapidly growing body of evidence. The spread of the COVID-19 pandemic requires the swift implementation of observational research to unravel heterogeneity of the natural course of the disease of SARS-CoV-2 infection in mechanically ventilated patients. Our study design is expected to enhance aetiological, diagnostic and prognostic understanding of the disease. This paper describes the design of the MaastrICCht. ETHICS AND DISSEMINATION: Ethical approval has been obtained from the medical ethics committee (Medisch Ethische Toetsingscommissie 2020-1565/3 00 523) of the Maastricht University Medical Centre+ (Maastricht UMC+), which will be performed based on the Declaration of Helsinki. During the pandemic, the board of directors of Maastricht UMC+ adopted a policy to inform patients and ask their consent to use the collected data and to store serum samples for COVID-19 research purposes. All study documentation will be stored securely for fifteen years after recruitment of the last patient. The results will be published in peer-reviewed academic journals, with a preference for open access journals, while particularly considering deposition of the manuscripts on a preprint server early. TRIAL REGISTRATION NUMBER: The Netherlands Trial Register (NL8613).

Improving Prediction of Favourable Outcome After 6 Months in Patients with Severe Traumatic Brain Injury Using Physiological Cerebral Parameters in a Multivariable Logistic Regression Model.

BACKGROUND/OBJECTIVE: Current severe traumatic brain injury (TBI) outcome prediction models calculate the chance of unfavourable outcome after 6 months based on parameters measured at admission. We aimed to improve current models with the addition of continuously measured neuromonitoring data within the first 24 h after intensive care unit neuromonitoring. METHODS: Forty-five severe TBI patients with intracranial pressure/cerebral perfusion pressure monitoring from two teaching hospitals covering the period May 2012 to January 2019 were analysed. Fourteen high-frequency physiological parameters were selected over multiple time periods after the start of neuromonitoring (0-6 h, 0-12 h, 0-18 h, 0-24 h). Besides systemic physiological parameters and extended Corticosteroid Randomisation after Significant Head Injury (CRASH) score, we added estimates of (dynamic) cerebral volume, cerebral compliance and cerebrovascular pressure reactivity indices to the model. A logistic regression model was trained for each time period on selected parameters to predict outcome after 6 months. The parameters were selected using forward feature selection. Each model was validated by leave-one-out cross-validation. RESULTS: A logistic regression model using CRASH as the sole parameter resulted in an area under the curve (AUC) of 0.76. For each time period, an increased AUC was found using up to 5 additional parameters. The highest AUC (0.90) was found for the 0-6 h period using 5 parameters that describe mean arterial blood pressure and physiological cerebral indices. CONCLUSIONS: Current TBI outcome prediction models can be improved by the addition of neuromonitoring bedside parameters measured continuously within the first 24 h after the start of neuromonitoring. As these factors might be modifiable by treatment during the admission, testing in a larger (multicenter) data set is warranted.

Improving long QT syndrome diagnosis by a polynomial-based T-wave morphology characterization.

BACKGROUND: Diagnosing long QT syndrome (LQTS) remains challenging because of a considerable overlap in QT interval between patients with LQTS and healthy subjects. Characterizing T-wave morphology might improve LQTS diagnosis. OBJECTIVE: The purpose of this study was to improve LQTS diagnosis by combining new polynomial-based T-wave morphology parameters with the corrected QT interval (QTc), age, and sex in a model. METHODS: A retrospective cohort consisting of 333 patients with LQTS and 345 genotype-negative family members was used in this study. For each patient, a linear combination of the first 2 Hermite-Gauss (HG) polynomials was fitted to the STT segments of an average complex of all precordial leads and limb leads I and II. The weight coefficients as well as the error of the best fit were used to characterize T-wave morphology. Subjects were classified as patients with LQTS or controls by clinical QTc cutoffs and 3 support vector machine models fed with different features. An external cohort consisting of 72 patients and 45 controls was finally used to check the robustness of the models. RESULTS: Baseline QTc cutoffs were specific but had low sensitivity in diagnosing LQTS. The model with T-wave morphology features, QTc, age, and sex had the best overall accuracy (84%), followed by a model with QTc, age, and sex (79%). The model with T-wave morphology features especially performed better in LQTS type 3 patients (69%). CONCLUSION: T-wave morphologies can be characterized by fitting a linear combination of the first 2 Hermite-Gauss polynomials. Adding T-wave morphology characterization to age, sex, and QTc in a support vector machine model improves LQTS diagnosis.

Artifacts in pulse transit time measurements using standard patient monitoring equipment.

OBJECTIVE: Pulse transit time (PTT) refers to the time it takes a pulse wave to travel between two arterial sites. PTT can be estimated, amongst others, using the electrocardiogram (ECG) and photoplethysmogram (PPG). Because we observed a sawtooth artifact in the PTT while using standard patient monitoring equipment for ECG and PPG, we explored the reasons for this artifact. METHODS: PPG and ECG were simulated at a heartrate of both 100 and 160 beats per minute while using a Masimo PPG post-processing module and a Philips patient monitor setup at the neonatal intensive care unit. Two different post-processing modules were used. PTT was defined as the difference between the R-peak in the ECG and the point of 50% increase in the PPG. RESULTS: A sawtooth artifact was seen in all simulations. Both length (59.2 to 72.4 s) and amplitude (30.8 to 36.0 ms) of the sawtooth were dependent on the post-processing module used. Furthermore, the absolute PTT value differed up to 250 ms depending on post-processing module and heart rate. The sawtooth occurred because the PPG wave continuously showed a minimal prolongation during the length of the sawtooth, followed by a sudden shortening. Both artifacts were generated in the post-processing module containing Masimo algorithms. CONCLUSION: Post-processing of the PPG signal in the Masimo module of the Philips patient monitor introduces a sawtooth in PPG and derived PTT. This sawtooth, together with a large module-dependent absolute difference in PTT, renders the thus-derived PTT insufficient for clinical purposes.

Support vector machine-based assessment of the T-wave morphology improves long QT syndrome diagnosis.

AIMS: Diagnosing long QT syndrome (LQTS) is challenging due to a considerable overlap of the QTc-interval between LQTS patients and healthy controls. The aim of this study was to investigate the added value of T-wave morphology markers obtained from 12-lead electrocardiograms (ECGs) in diagnosing LQTS in a large cohort of gene-positive LQTS patients and gene-negative family members using a support vector machine. METHODS AND RESULTS: A retrospective study was performed including 688 digital 12-lead ECGs recorded from genotype-positive LQTS patients and genotype-negative relatives at their first visit. Two models were trained and tested equally: a baseline model with age, gender, RR-interval, QT-interval, and QTc-intervals as inputs and an extended model including morphology features as well. The best performing baseline model showed an area under the receiver-operating characteristic curve (AUC) of 0.821, whereas the extended model showed an AUC of 0.901. Sensitivity and specificity at the maximal Youden's indexes changed from 0.694 and 0.829 with the baseline model to 0.820 and 0.861 with the extended model. Compared with clinically used QTc-interval cut-off values (>480 ms), the extended model showed a major drop in false negative classifications of LQTS patients. CONCLUSION: The support vector machine-based extended model with T-wave morphology markers resulted in a major rise in sensitivity and specificity at the maximal Youden's index. From this, it can be concluded that T-wave morphology assessment has an added value in the diagnosis of LQTS.

The development and validation of an easy to use automatic QT-interval algorithm.

BACKGROUND: To evaluate QT-interval dynamics in patients and in drug safety analysis, beat-to-beat QT-interval measurements are increasingly used. However, interobserver differences, aberrant T-wave morphologies and changes in heart axis might hamper accurate QT-interval measurements. OBJECTIVE: To develop and validate a QT-interval algorithm robust to heart axis orientation and T-wave morphology that can be applied on a beat-to-beat basis. METHODS: Additionally to standard ECG leads, the root mean square (ECGRMS), standard deviation and vectorcardiogram were used. QRS-onset was defined from the ECGRMS. T-wave end was defined per individual lead and scalar ECG using an automated tangent method. A median of all T-wave ends was used as the general T-wave end per beat. Supine-standing tests of 73 patients with Long-QT syndrome (LQTS) and 54 controls were used because they have wide ranges of RR and QT-intervals as well as changes in T-wave morphology and heart axis orientation. For each subject, automatically estimated QT-intervals in three random complexes chosen from the low, middle and high RR range, were compared with manually measured QT-intervals by three observers. RESULTS: After visual inspection of the randomly selected complexes, 21 complexes were excluded because of evident noise, too flat T-waves or premature ventricular beats. Bland-Altman analyses of automatically and manually determined QT-intervals showed a bias of <4ms and limits of agreement of ±25ms. Intra-class coefficient indicated excellent agreement (>0.9) between the algorithm and all observers individually as well as between the algorithm and the mean QT-interval of the observers. CONCLUSION: Our automated algorithm provides reliable beat-to-beat QT-interval assessment, robust to heart axis and T-wave morphology.