Yair Goldberg's Lab

Our group develops statistical and machine-learning methods for complex data, with applications in health, medicine and public policy. We work with researchers and public institutions to turn data into reliable evidence and better decisions.

Research

Methodological work developed alongside applications in public health, clinical medicine and health-care policy.

Mental Health and Healthy Ageing

Dementia risk reflects changing health, treatments and social conditions from midlife onwards. We develop methods that update risk as new information accumulates.

Learn more

From longitudinal records to dynamic prevention

Cognitive decline and dementia develop over many years, while the information available about an individual changes continuously. Health conditions, medications, laboratory results, functional measures and social circumstances may all affect risk, and their relevance can vary over time.

Our work develops methods for analysing these complex longitudinal data and producing updated, individual-level risk predictions. This makes it possible to identify periods in which prevention, monitoring or clinical support may be most valuable.

Related work

Rotstein A, Kodesh A, Goldberg Y, et al. Serum folate deficiency and the risks of dementia and all-cause mortality. Evidence-Based Mental Health, 2022.
Travis-Lumer Y, Kodesh A, Goldberg Y, et al. Attempted suicide rates before and during the COVID-19 pandemic. Psychological Medicine, 2023.
Travis-Lumer Y, Kodesh A, Goldberg Y, et al. Biopsychosocial exposure to the COVID-19 pandemic and the relative risk of schizophrenia. European Psychiatry, 2022.
Frangou S, Travis-Lumer Y, Kodesh A, Goldberg Y, et al. Increased incident rates of antidepressant use during the COVID-19 pandemic. Psychological Medicine, 2023.

Abstract health-care capacity and queues

Health-care Access and Waiting Times

We study how demand, referrals and service capacity interact to shape access to care. Additional capacity alone does not necessarily reduce waiting times: demand and referral patterns may change as services expand.

Learn more

Measuring access, understanding queues, improving care

Waiting time is not simply a measure of how many appointments are available. It reflects a dynamic system: patients enter through different referral pathways, providers differ in capacity and practice, and demand changes in response to clinical need, policy and service availability.

We use population-level administrative and clinical data to measure waiting times across services and patient groups, identify disparities, and evaluate changes in policy or capacity. The goal is to provide evidence that helps health systems decide where interventions are most likely to improve timely care.

Related work

Murad H, Myers V, Ziv A, Wilf-Miron R, Goldberg Y, et al. Measuring geographical disparities in waiting times for community-based specialist care. Israel Journal of Health Policy Research, 2025.
Waiting times for elective surgical procedures: Implementation of a national measurement methodology. Manuscript in preparation.
Smart Learning or Simple Rules? MRI Capacity Planning with Endogenous Demand. Manuscript submitted.
Silent Abandonment in Text-Based Contact Centers: Identifying, Quantifying, and Mitigating its Operational Impacts. Manuscript submitted.

Abstract vaccination protection over time

Vaccines and Infectious Disease

Using nationwide health data, we study vaccine effectiveness, waning immunity and the benefit of booster doses as variants, prior infection and population behaviour change over time.

Learn more

Evidence that evolves with the epidemic

The effect of a vaccine is not fixed. Protection can change with time since vaccination, the emergence of new variants, prior infection, age, exposure patterns and the evolving composition of the population at risk.

During the COVID-19 pandemic, our group worked with nationwide Israeli data to estimate vaccine effectiveness and its waning over time, and to assess the added protection provided by booster doses. This work continues more broadly in infectious diseases, supporting vaccination policy as new evidence accumulates.

Selected publications

Goldberg Y, Mandel M, Bar-On YM, et al. Waning Immunity after the BNT162b2 Vaccine in Israel. NEJM, 2021.
Bar-On YM, Goldberg Y, Mandel M, et al. Protection of BNT162b2 Vaccine Booster against Covid-19 in Israel. NEJM, 2021.
Goldberg Y, Mandel M, Bar-On YM, et al. Protection and Waning of Natural and Hybrid Immunity to SARS-CoV-2. NEJM, 2022.
Bar-On YM, Goldberg Y, Mandel M, et al. Protection by a Fourth Dose of BNT162b2 against Omicron in Israel. NEJM, 2022.
Goldberg Y, Huppert A. To boost or not to boost: navigating post-pandemic COVID-19 vaccination. The Lancet Respiratory Medicine, 2023.

Abstract hereditary breast cancer prevention

Precision Prevention of Hereditary Breast Cancer

We develop a national, life-course approach to genetic testing, screening and preventive care for women at hereditary risk of breast and ovarian cancer.

Learn more

From a genetic result to lifelong prevention

A pathogenic BRCA1 or BRCA2 variant has important implications for cancer prevention, but it is not a complete prevention plan. Risk and the appropriate preventive options change with age, family history, screening findings, reproductive history, comorbidities and personal preferences.

We are developing a national research platform that follows women from genetic testing through screening, preventive choices and cancer outcomes. The aim is to understand how prevention unfolds in real life, where gaps and inequities arise, and which strategies are most effective for different groups of women.

Autoimmune Disease Flares

We use longitudinal health records, laboratory results and treatment histories to estimate recurrent flare risk and support earlier, more informed care.

Learn more

From fragmented records to proactive care

Autoimmune diseases often follow a relapsing-remitting course, but a flare is rarely recorded as one clear, standardised event. We develop reproducible ways to identify flares from medication changes, urgent contacts, emergency visits and hospitalisations, inflammatory laboratory markers, and clinical follow-up.

We then model time to the next flare as a recurrent time-to-event problem, updating risk as new information becomes available. We work with longitudinal data from hospitals, health maintenance organizations and the Kinneret Israel Health Data Lake to study disease activity and recurrent clinical events across autoimmune conditions.

Related work

Vadasz Z, Goldberg Y, et al. Increased soluble CD72 in systemic lupus erythematosus is in association with disease activity and lupus nephritis. Clinical Immunology, 2016.
Vadasz Z, Goldberg Y, et al. Lysyl oxidase—a possible role in systemic sclerosis–associated pulmonary hypertension: a multicentre study. Rheumatology, 2019.
Willner N, Goldberg Y, Schiff E, Vadasz Z. Semaphorin 4D levels in heart failure patients: a potential novel biomarker of acute heart failure? ESC Heart Failure, 2018.

Abstract diagram of missing and censored data

Machine Learning for Missing and Censored Data

We develop machine-learning tools for missing covariates and responses, right-censoring, left truncation and other forms of incomplete or selectively observed data.

Learn more

Learning from incomplete real-world data

Machine-learning methods are often developed under the assumption that the data are complete. In real applications, covariates or outcomes may be missing; an outcome may only be known to exceed a threshold; individuals may enter the data only after reaching a given condition; or a label may be available only for a group of observations.

Our work develops methods that remain reliable under these forms of incomplete information. We combine imputation, inverse-probability weighting and doubly robust estimation with flexible kernel-based and high-dimensional learning methods, providing principled uncertainty quantification and clearer guidance on when a prediction can be trusted.

Selected work