Steffen Albrecht

Summary / Background: Models based on administrative population health records can help improve understanding of disease and hospitalisation risk. This understanding can support evidence-based population health policy and processes. Influenza and other acute respiratory diseases place a large, inequitable burden on the New Zealand population and health care resources. We have developed and evaluated models using retrospective New Zealand hospitalisation records linked to other administrative data. Methods: We analysed linked annual records for approximately 4·85–5·36 million people in 2017–2024. For each year, models were trained to distinguish people who experienced at least one hospitalisation associated with respiratory infectious disease from those without such a hospitalisation during that year. Six algorithms and eight variable sets were assessed using five-fold cross-validation grouped by demographic profile. Variables comprised demographics, continuous hospital-based (M3) and pharmaceutical-based (P3) morbidity indices, and influenza vaccination status. Models trained on one year were also evaluated on data from the following year. Performance measures were the area under the receiver operating characteristic curve (AUROC) and the average precision score (APS). SHapley Additive exPlanations (SHAP) described model contributions, including paired vaccination-input changes in 2024. Findings: Annual national hospitalisation prevalence ranged from 1·01% to 1·60%. With all variables, mean cross-validation AUROC and APS were 0·8430 and 0·1283 for LightGBM (LGB), and 0·8415 and 0·1285 for XGBoost (XGB). Adding both morbidity indices to the demographics increased LGB APS from 0·0720 to 0·1158; adding vaccination further increased it to 0·1283. Next-year AUROC and APS averaged 0·8429 and 0·1267 for LGB and 0·8423 and 0·1277 for XGB. Age and morbidity were the most influential variables. Among vaccinated individuals, SHAP analysis indicated that vaccination contributed most strongly towards lower model-estimated hospitalisation risk in the oldest age group. Interpretation: Combining historical morbidity and vaccination information in gradient boosting models improved the ability to distinguish between people who experienced respiratory hospitalisation and those who did not in annual New Zealand population data. Similar next-year performance and interpretable variation across age and morbidity profiles support further development toward practical risk assessment, including calibration and evaluation across population groups. Research in context: Evidence before this study: Our focused narrative review considered population-based respiratory hospitalisation models, respiratory disease burden in New Zealand, and the development of the M3 and P3 morbidity indices. Existing studies include population models for COVID-19 and infant respiratory syncytial virus, risk-factor analyses of broad respiratory admissions, and evidence that hospital and pharmaceutical morbidity measures provide complementary information. Added value of this study: This is the first evaluation of national hospitalisation data linked to other national administrative data to model the risk of acute respiratory infection admission across eight annual datasets, comparing six algorithms and eight variable sets. Together with the benchmark results, these findings illustrate how gradient boosting applied to population-scale records can represent and explain combinations of routinely recorded health information. Implications of all the available evidence: Quantifying historical morbidity and including vaccination information improved respiratory hospitalisation classification, with gradient boosting providing the strongest overall performance. The explanatory analyses further showed that vaccination-related contributions depended on the individual’s age and morbidity profile. Representing these clinically plausible patterns supports the use of flexible, non-linear models, such as the tree-based algorithms, and rich population-scale records for developing interpretable respiratory risk assessments. The study establishes a basis for a New Zealand respiratory risk model by identifying effective algorithms and useful combinations of variables.

Introduction Influenza is a major global health concern, responsible for up to 650 000 respiratory-related deaths annually. Although influenza is often perceived as mild in healthy adults, it can cause severe outcomes in high-risk groups, such as older adults, young children, pregnant women and those with underlying medical conditions. Various clinical, sociodemographic and environmental factors influence the progression to severe outcomes, whereas resilience factors, such as vaccination, may reduce risks. Despite growing research, the evidence base regarding risk and resilience is spread across many different aspects of the literature. This umbrella review will synthesise evidence from existing systematic reviews and meta-analyses to identify key risk and resilience factors associated with the progression of influenza to severe outcomes in the general population. Methods and analysis This umbrella review follows the Joanna Briggs Institute (JBI) and Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols guidelines. We will include systematic reviews and meta-analyses reporting host-related risk or resilience factors for severe influenza outcomes. Four databases (EMBASE, Scopus, Medline and CINAHL) will be searched for English-language publications. Study quality will be assessed using AMSTAR 2, and the body of evidence will be evaluated using Grading of Recommendations Assessment, Development and Evaluation. Due to heterogeneity, findings will be analysed narratively. Risk and resilience factors will be grouped into demographic, clinical, behavioural, social and psychological domains. Ethics and disseminations No ethical approval is required. The completed review will be shared through peer-reviewed journals and conference presentations. PROSPERO registration number CRD420250644475.

Background: Public health surveillance systems need to monitor influenza activity and guide measures to mitigate its high impact on morbidity, mortality and healthcare systems. There is an increasing expectation that surveillance data will support the modeling of future short‐term disease scenarios using artificial intelligence (AI) and machine learning (ML). This study examines how influenza surveillance can support AI/ML‐based short‐term forecasting for influenza at the community and hospital levels in a high‐income country setting (Aotearoa/New Zealand).Methods: This study used a two‐phase approach. The first phase involved a comprehensive review of government reports, official websites, and literature to characterize existing influenza surveillance systems. The second phase evaluated systems against eight key attributes—timeliness, sensitivity, specificity, representativeness, coverage, robustness, completeness, and historical data—using a five‐level ranking system. Attribute selection was informed by experts' knowledge, ML requirements, and established frameworks. Weighted scores for training and short‐term forecasting capabilities were calculated to determine alignment with AI/ML requirements.Results: The Southern Hemisphere Influenza and Vaccine Effectiveness Research and Surveillance (SHIVERS) community cohort and Severe Acute Respiratory Infection (SARI) hospital surveillance emerged as the most useful systems, achieving the highest scores in both training and short‐term forecasting in community and hospital settings, respectively. The National Minimum Dataset of hospitalizations and mortality datasets demonstrated strong training potential but are limited in short‐term forecasting due to timeliness constraints. Additionally, laboratory‐based surveillance performs a useful role in bridging community and hospital datasets.Conclusions: A set of key attributes is useful for assessing which influenza surveillance systems are best aligned with AI/ML training and short‐term forecasting requirements. These attributes distinguished systems that are likely to be the most suitable for modeling future short‐term disease scenarios for influenza at the community and hospital levels in New Zealand. Integrating these data sources could enhance influenza forecasts to improve public health responses and intervention planning.

This study explores the genetic regulators of microRNAs (miRNAs) using an ensemble of machine learning models to predict miRNA expression levels from gene expression data. Employing ridge regression, we accurately predicted the expression of 353 human miRNAs (R ² > 0.5), revealing robust miRNA-gene regulatory relationships. By analyzing the coefficients of these predictive models, we identified genetic regulators for each miRNA and highlighted the multifactorial nature of miRNA regulation. Further network analysis uncovered that miRNAs with higher predictive accuracy are more densely connected to their top predictive genes, reflecting strong regulatory control within miRNA-gene networks. To refine these insights, we filtered the gene-miRNA interaction networks to identify miRNAs specifically associated with enriched pathways, such as synaptic function and cardiovascular processes. From this pathway-centric analysis, we present a curated list of miRNAs and their genetic regulators, pinpointing their activity within distinct biological contexts. Additionally, our study provides a comprehensive set of metrics and coefficients for the genes most predictive of miRNA expression, along with a filtered subnetwork of miRNAs linked to specific pathways and phenotypes. By integrating miRNA expression predictors with network analysis and pathway enrichment, this work advances our understanding of miRNA regulatory mechanisms and their roles across distinct biological systems.

With the help of high-performance computing, we benchmarked a selection of machine learning classification algorithms on the tasks of whisker stimulus detection, stimulus classification and behavior prediction based on electrophysiological recordings of layer-resolved local field potentials from the barrel cortex of awake mice. Machine learning models capable of accurately analyzing and interpreting the neuronal activity of awake animals during a behavioral experiment are promising for neural prostheses aimed at restoring a certain functionality of the brain for patients suffering from a severe brain injury. The liquid state machine, a highly efficient spiking neural network classifier that was designed for implementation on neuromorphic hardware, achieved the same level of accuracy compared to the other classifiers included in our benchmark study. Based on application scenarios related to the barrel cortex and relevant for neuroprosthetics, we show that the liquid state machine is able to find patterns in the recordings that are not only highly predictive but, more importantly, generalizable to data from individuals not used in the model training process. The generalizability of such models makes it possible to train a model on data obtained from one or more individuals without any brain lesion and transfer this model to a prosthesis required by the patient. Author Summary A neural prosthesis is a computationally driven device that restores the functionality of a damaged brain region for locked-in patients suffering from the aftereffects of a brain injury or severe stroke. As such devices are chronically implanted, they rely on small, low-powered microchips with limited computational resources. Based on recordings describing the neural activity of awake mice, we show that spiking neural networks, which are especially designed for microchips, are able to provide accurate classification models in application scenarios relevant in neuroprosthetics. Furthermore, models were generalizable across mice, corroborating that it will be possible to train a model on recordings from healthy individuals and transfer it to the patient’s prosthesis.

Motivation: Single-cell Chromatin ImmunoPrecipitation DNA-Sequencing (scChIP-seq) analysis is challenging due to data sparsity. High degree of sparsity in biological high-throughput single-cell data is generally handled with imputation methods that complete the data, but specific methods for scChIP-seq are lacking. We present SIMPA, a scChIP-seq data imputation method leveraging predictive information within bulk data from the ENCODE project to impute missing protein-DNA interacting regions of target histone marks or transcription factors.Results: Imputations using machine learning models trained for each single cell, each ChIP protein target, and each genomic region accurately preserve cell type clustering and improve pathway-related gene identification on real human data. Results on bulk data simulating single cells show that the imputations are single-cell specific as the imputed profiles are closer to the simulated cell than to other cells related to the same ChIP protein target and the same cell type. Simulations also show that 100 input genomic regions are already enough to train single-cell specific models for the imputation of thousands of undetected regions. Furthermore, SIMPA enables the interpretation of machine learning models by revealing interaction sites of a given single cell that are most important for the imputation model trained for a specific genomic region. The corresponding feature importance values derived from promoter-interaction profiles of H3K4me3, an activating histone mark, highly correlate with co-expression of genes that are present within the cell-type specific pathways in 2 real human and mouse datasets. The SIMPA’s interpretable imputation method allows users to gain a deep understanding of individual cells and, consequently, of sparse scChIP-seq datasets.Availability and implementation: Our interpretable imputation algorithm was implemented in Python and is available athttps://github.com/salbrec/SIMPA.

Classification problems in the small data regime (with small data statistic T and relatively large feature space dimension D) impose challenges for the common machine learning (ML) and deep learning (DL) tools. The standard learning methods from these areas tend to show a lack of robustness when applied to data sets with significantly fewer data points than dimensions and quickly reach the overfitting bound, thus leading to poor performance beyond the training set. To tackle this issue, we propose eSPA+, a significant extension of the recently formulated entropy-optimal scalable probabilistic approximation algorithm (eSPA). Specifically, we propose to change the order of the optimization steps and replace the most computationally expensive subproblem of eSPA with its closed-form solution. We prove that with these two enhancements, eSPA+ moves from the polynomial to the linear class of complexity scaling algorithms. On several small data learning benchmarks, we show that the eSPA+ algorithm achieves a many-fold speed-up with respect to eSPA and even better performance results when compared to a wide array of ML and DL tools. In particular, we benchmark eSPA+ against the standard eSPA and the main classes of common learning algorithms in the small data regime: various forms of support vector machines, random forests, and long short-term memory algorithms. In all the considered applications, the common learning methods and eSPA are markedly outperformed by eSPA+, which achieves significantly higher prediction accuracy with an orders-of-magnitude lower computational cost.

To address the question which neocortical layers and cell types are important for the perception of a sensory stimulus, we performed multielectrode recordings in the barrel cortex of head-fixed mice performing a single-whisker go/no-go detection task with vibrotactile stimuli of differing intensities. We found that behavioral detection probability decreased gradually over the course of each session, which was well explained by a signal detection theory-based model that posits stable psychometric sensitivity and a variable decision criterion updated after each reinforcement, reflecting decreasing motivation. Analysis of multiunit activity demonstrated highest neurometric sensitivity in layer 4, which was achieved within only 30 ms after stimulus onset. At the level of single neurons, we observed substantial heterogeneity of neurometric sensitivity within and across layers, ranging from nonresponsiveness to approaching or even exceeding psychometric sensitivity. In all cortical layers, putative inhibitory interneurons on average proffered higher neurometric sensitivity than putative excitatory neurons. In infragranular layers, neurons increasing firing rate in response to stimulation featured higher sensitivities than neurons decreasing firing rate. Offline machine-learning-based analysis of videos of behavioral sessions showed that mice performed better when not moving, which at the neuronal level, was reflected by increased stimulus-evoked firing rates.