Research

Economic evaluations often rely on observational data, in which treated and untreated patients differ in ways that affect outcomes. An early strand of my work compared estimators for this setting. Using simulations and applied case studies, we found that semiparametric methods such as targeted maximum likelihood estimation and bias-corrected matching performed better than standard propensity score approaches, in terms of bias and efficiency, when the models were misspecified.

Later work extended these methods to longitudinal settings with time-varying confounding, including the evaluation of dynamic treatment regimes in critical care, and to electronic health record data in which biomarkers are measured only when clinicians decide to measure them. The common aim has been to make adaptive treatment strategies evaluable with the data that health systems routinely collect.

Key papers

• Kreif N, Sofrygin O, Schmittdiel JA, Adams AS, Grant RW, Zhu Z, van der Laan MJ, Neugebauer R (2021). Exploiting non-systematic covariate monitoring to broaden the scope of evidence about the causal effects of adaptive treatment strategies. Biometrics, 77(1), 329–342.
• Kreif N, Tran L, Grieve R, De Stavola B, Tasker RC, Petersen M (2017). Estimating the comparative effectiveness of feeding interventions in the pediatric intensive care unit: a demonstration of longitudinal targeted maximum likelihood estimation. American Journal of Epidemiology, 186(12), 1370–1379.
• Kreif N, Gruber S, Radice R, Grieve R, Sekhon JS (2016). Evaluating treatment effectiveness under model misspecification: a comparison of targeted maximum likelihood estimation with bias-corrected matching. Statistical Methods in Medical Research, 25(5), 2315–2336.
• Kreif N, Grieve R, Sadique MZ (2013). Statistical methods for cost-effectiveness analyses that use observational data: a critical appraisal tool and review of current practice. Health Economics, 22(4), 486–500.
• Kreif N, Grieve R, Radice R, Sadique Z, Ramsahai R, Sekhon JS (2012). Methods for estimating subgroup effects in cost-effectiveness analyses that use observational data. Medical Decision Making, 32(6), 750–763.

A policy with a small average effect may help some groups and not others. This strand applies causal machine learning methods, including causal forests, meta-learners and policy learning, to estimate heterogeneous treatment effects and to learn targeting rules in health policy and clinical settings. Applications include Indonesia's national health insurance program and the question of how individualized treatment effect estimates can be used within cost-effectiveness decision models.

On the methods side, an MRC New Investigator Research Grant with James Cussens supported work on policy learning with rare outcomes, algorithms for learning optimal policy trees, and machine learning for staggered difference-in-differences designs. I was a member of the ISPOR task force that produced the PALISADE checklist for machine learning in health economics and outcomes research, and co-led a review of methods for estimating individualized treatment effects for use in health technology assessment.

Key papers

• Shah V^‡, Jones AM, Malenica I, Hidayat T, Kreif N (2025). Using policy learning to inform health insurance targeting: a case study of Indonesia. Health Economics, 34(12), 2270–2296.
• Zhang Y, Kreif N, GC VS, Manca A (2024). Machine learning methods to estimate individualized treatment effects for use in health technology assessment. Medical Decision Making, 44(7), 756–769.
• Glynn D, Giardina J, Hatamyar J^‡, Pandya A, Soares M, Kreif N (2024). Integrating decision modeling and machine learning to inform treatment stratification. Health Economics, 33(8), 1772–1792.
• Hatamyar J^‡, Kreif N (2023). Policy learning with rare outcomes. arXiv:2302.05260.
• Padula WV, Kreif N, Vanness DJ, Adamson B, Rueda JD, Felizzi F, Jonsson P, IJzerman MJ, Butte A, Crown W (2022). Machine learning methods in health economics and outcomes research — the PALISADE checklist: a good practices report of an ISPOR task force. Value in Health, 25(7), 1063–1080.
• Kreif N, DiazOrdaz K, Moreno-Serra R, Mirelman A, Hidayat T, Suhrcke M (2022). Estimating heterogeneous policy impacts using causal machine learning: a case study of health insurance reform in Indonesia. Health Services and Outcomes Research Methodology, 22(2), 192–227.

When a payment reform or an incentive scheme is introduced across a whole health system, the counterfactual has to be constructed from comparison units. Using data from the English NHS, we examined how the synthetic control method behaves when there are multiple treated units, and compared it with difference-in-differences and related designs, to show how conclusions can depend on the choice of method.

We have used these designs to evaluate hospital payment reforms and provider incentive programs in England. More recent work, with colleagues at Imperial College London, evaluates digital-first models of primary care and the trade-off between equity and efficiency in access.

Key papers

• Rodrigues D^‡, Kreif N, Darzi A, Barahona M, Mayer E (2025). Digitalisation of access to primary care: is there an equity-efficiency trade-off?. Health Economics, 34(10), 1943–1962.
• Aragón MJ, Chalkley M, Kreif N (2022). The long-run effects of diagnosis related group payment on hospital lengths of stay in a publicly funded health care system: evidence from 15 years of micro data. Health Economics, 31(6), 956–972.
• Rodrigues D^‡, Kreif N, Saravanakumar K, Delaney B, Barahona M, Mayer E (2022). Formalising triage in general practice towards a more equitable, safe, and efficient allocation of resources. BMJ, 377, e070757.
• O'Neill S, Kreif N, Sutton M, Grieve R (2020). A comparison of methods for health policy evaluation with controlled pre-post designs. Health Services Research, 55(2), 328–338.
• Gaughan J, Gutacker N, Grašič K, Kreif N, Siciliani L, Street A (2019). Paying for efficiency: incentivising same-day discharges in the English NHS. Journal of Health Economics, 68, 102226.
• Kreif N, Grieve R, Hangartner D, Turner AJ, Nikolova S, Sutton M (2016). Examination of the synthetic control method for evaluating health policies with multiple treated units. Health Economics, 25(12), 1514–1528.

This strand applies causal methods to questions of health system performance in low- and middle-income countries, including quality improvement policies, health expenditure expansions and provider payment reforms. In South Africa, we examined whether national facility quality standards improved care uniformly, or only for clinics already above a baseline level of quality, which has implications for how quality incentives are designed and targeted.

In Indonesia, we estimated the productivity of the health system, and these estimates have been used in supply-side cost-effectiveness threshold estimates for reimbursement decisions. Related work covers maternal health care in Malawi. Much of this research was carried out under the NIHR-funded GHE2 program at the University of York and the NIHR Global Health Research Unit on health financing for universal health coverage.

Key papers

• Ochoa-Moreno I, Kreif N, Hidayat T, Kurnia S, Hidayat B, Griffin S (2026). Does increasing overall health expenditure reduce inequality in under-5 mortality rates between provinces in Indonesia?. Health Policy and Planning, 41(7), 1242–1254.
• McGuire F^‡, Smith PC, Stacey N, Edoka I, Kreif N (2025). Do health care quality improvement policies work for all? Distributional effects by baseline quality in South Africa. Health Economics, 34(1), 175–199.
• Moler-Zapata S^‡, Kreif N, Ochalek J, Mirelman AJ, Nadjib M, Suhrcke M (2022). Estimating the health effects of expansions in health expenditure in Indonesia: a dynamic panel data approach. Applied Health Economics and Health Policy, 20(6), 881–891.
• McGuire F^‡, Kreif N, Smith PC (2021). The effect of distance on maternal institutional delivery choice: evidence from Malawi. Health Economics, 30(9), 2144–2167.
• Stacey N, Mirelman A, Kreif N, Suhrcke M, Hofman K, Edoka I (2021). Facility standards and the quality of public sector primary care: evidence from South Africa's "Ideal Clinics". Health Economics, 30(7), 1543–1558.

Conflict, violence and mental ill-health do not occur at random, which makes their consequences hard to measure. This strand applies causal inference methods to these questions. With Rodrigo Moreno-Serra and colleagues at York, we studied the effects of armed conflict in Colombia on child health and on the financing of the health system.

In the United States, we estimated the effect of psychological distress in adolescence on health and economic outcomes in adulthood. In Brazil, the BRAVE project examined the relationship between pandemic distancing policies and violence against women. My current MQ Foundation Fellowship in Economics and Mental Health Research continues this work on the economics of mental health.

Key papers

• Diaz MD, Pereda P, Rocha F, Oliveira P, Árabe I, Kreif N, Lordemus S, Moreno-Serra R (2025). Public policies and femicides during the COVID-19 pandemic: evidence from São Paulo, Brazil. Economics & Human Biology, 58, 101508.
• Counts NZ, Kreif N, Creedon TB, Bloom DE (2025). Psychological distress in adolescence and later economic and health outcomes in the US population. PLOS Medicine, 22(1), e1004506.
• Lordemus S, Kreif N, Moreno-Serra R (2024). Public healthcare financing during counterinsurgency efforts: evidence from Colombia. Oxford Bulletin of Economics and Statistics, 86(5), 1230–1259.
• Rocha F, Diaz MDM, Pereda PC, Árabe IB, Cavalcanti F, Lordemus S, Kreif N, Moreno-Serra R (2024). COVID-19 and violence against women: current knowledge, gaps, and implications for public policy. World Development, 174, 106461.
• Kreif N, Mirelman A, Suhrcke M, Buitrago G, Moreno-Serra R (2022). The impact of civil conflict on child health: evidence from Colombia. Economics & Human Biology, 44, 101074.

fastpolicytree (R package, CRAN). Constructs optimal policy trees from covariate data and doubly robust reward estimates. It is faster than the policytree package, particularly for discrete covariates. James Cussens developed it within the MRC-funded Tailoring project, which I led; the methods are described in Cussens, Hatamyar, Shah and Kreif (2025). Install with install.packages("fastpolicytree").

I was a visiting graduate student at the University of California, Berkeley, in 2010, a visiting researcher in its Division of Biostatistics in 2015, and a visiting researcher at its Center for Targeted Machine Learning and Causal Inference in 2022. In 2016 I was a visiting researcher at the Kaiser Permanente Division of Research.

I work with economists, biostatisticians, epidemiologists and clinicians at the University of Washington, the University of York, the London School of Hygiene and Tropical Medicine, Imperial College London and Harvard, and with partners in Indonesia, Colombia, Brazil, South Africa and Malawi.