A 2-Day Workshop by Professor Charles Manski
This workshop is geared to PhDs, post-doctoral fellows, researchers, clinicians, health economists, statisticians, and others interested in statistical methods in healthcare.
Dates: April 11 (10am to 5pm, followed by an optional dinner) and April 12 (10am to 3pm), 2018.
Location: McMaster Innovation Park, Hamilton, Ontario
To register: Click here (workshop: $190 non-students; $50 for students; optional dinner: $50 per person.)
The Centre for Health Economics and Policy Analysis (CHEPA), the Dept. of Economics, The Dept. of Health Research Methods, Evidence, and Impact (HEI), McMaster University, and the Canadian Centre for Health Economics (CCHE), University of Toronto, are proud to host Dr. Manski at McMaster University for a 2-day course focussed on cutting edge statistical/econometric approaches that can be used to enhance the development of clinical guidelines and in informing treatment choice.
Prof. Charles Manski pioneered advances in partial identification across many policy applications. His recent work addresses opportunities to build further robustness and rigour into clinical decision-making in relation to patient care under uncertainty with applications to clinical guideline development, prognosis and treatment.
Clinical research has favoured trial data because of the limited internal validity of observational studies, yet trials have methodological problems as well. Small trial sizes limit subgroup specific conclusions for personalized care decisions of multimorbid patients. Extrapolation from trials to clinical practice may be difficult for multiple reasons. Partial identification analysis can address this challenge by providing reliable information on the range of treatment effect sets that are consistent with observational or trial data by considering all credible configurations of counterfactuals (i.e., the possible values of outcomes under alternative treatments). Even wide ranges of estimates can be useful information in a decision problem.
Partial identification and clinical guidelines
Partial identification analysis also provides a robust way to enrich the information on risk factors used in prediction tools. For example, the Breast Cancer Risk Assessment (BCRA) tool of the National Cancer Institute provides evidence-based cancer risk assessment along eight recognized risk factors. But other pertinent factors such as presence of cancer among second-degree relatives or excessive drinking behaviour are not part of the statistical model used by the tool. Clinicians who make treatment choices using BCRA derived information may wish to consider how such factors should modify recommendations to personalize care according to each patient’s background. However, psychological studies of clinical decision-making suggest informal use of such extra information does not result in better treatment decisions on average.
Charles F. Manski is a superb speaker who can make technical information accessible to broad audiences. He is currently working on a book on Personalizing Patient Care Under Uncertainty. Previous books include Public Policy in an Uncertain World (Harvard 2013), Identification for Prediction and Decision (Harvard 2007), Social Choice with Partial Knowledge of Treatment Response (Princeton 2005), Partial Identification of Probability Distributions (Springer, 2003), Identification Problems in the Social Sciences (Harvard 1995), and Analog Estimation Methods in Econometrics (Chapman & Hall, 1988).
He is Board of Trustees Professor in Economics at Northwestern University. He is an elected member of the U.S. National Academy of Sciences, and Fellow of the: Econometric Society, American Statistical Association, American Academy of Arts and Sciences, American Association for the Advancement of Science, and the British Academy. More information on Dr. Manski can be found at: (https://en.wikipedia.org/wiki/Charles_F._Manski)
Clinical guidelines often rate treatment options by the degree of certainty in some magnitude of net benefit. Decision analytic methods can naturally formalize qualitative aspects of treatment recommendation in this setting. Systems of ranking evidence across studies such as GRADE also use qualitative criteria to rank studies in terms of design and potential bias. Partial identification analysis can help quantify some aspects of study robustness and validity to inform the treatment decision. For example, for trials with non-random attrition, partial identification analysis of treatment effects provides quantitative information on a study’s internal validity.
Statistical decision theory with partial knowledge of probability distributions is a robust, comprehensive, and innovative approach to prognosis and treatment choice in a frequentist paradigm. It does not require the parametric distributional assumptions of popular Bayesian methods. Neither does it make the assumptions of classical frequentist regression to point identify a parameter, nor rely on the arbitrary critical values of classical hypothesis testing. Instead, it formally specifies a normative component of the decision problem (e.g., a metric of patient welfare across health outcomes), and a predictive component (e.g., the ex ante likelihood of patient outcomes under a treatment regime). The predictive component implies uncertainty on two dimensions:
1) Uncertainty which decreases with sample size (the problem of sampling error), and
2) Uncertainty which does not decrease with sample size (the problem of identification).
This workshop will deal with the problem of identification focussing on clinical guideline development. Although the foundation of the methodology is technical, course content will be mostly conceptual and will emphasize the utility of the approaches for the clinical practitioner and those interested in the development of clinical guidelines. Basic statistical literacy is sufficient for participation. Technical questions will be accommodated as feasible.
- Review of psychological and economic research comparing evidence-based prediction with clinical judgment in personalized medicine
- Critical appraisal of inappropriate extrapolation of results from trials and meta-analysis, and limitations of hypothesis testing for treatment decisions and power-based approaches to choosing trial sample size
- Formal analysis of partial identification and extrapolation of treatment effects with experimental and observational data, including missing data problems
- Formal analysis of decision under uncertainty with experimental and observational data, including discussion of optimal and reasonable decision criteria; rational expectations and subjective utility-based decisions; and decisions under ambiguity
- Case study with course methods: nodal observation or dissection in treatment of melanoma
- An adaptive diversification approach to public health policy which holds promise for reducing uncertainty over time, while fulfilling reasonable decision criteria in the present
Examples and applications may include optimal testing and treatment, surveillance versus aggressive treatment, breast cancer risk assessment, cardiovascular disease prediction, bariatric surgery outcome assessment, and hypertension drug trial outcome assessment.
Frequently Asked Questions:
What are my transportation/parking options for getting to and from the event?
Visitor parking payment is required from 8:00 a.m. to 5:00 p.m. Monday to Friday. Exact change, Mastercard or Visa, are the payment options. All guests visiting Tenants in the building are directed to the Visitor Parking area.
VISITOR PARKING AREAS
- Lot 1, adjacent to the ATRIUM@MIP at 175 Longwood Road South, Hamilton, and is a pay-and-display parking area
- Lot 5, gated gravel lot located north of Lot 1
- South of the CanMet building on Frid Street
- Visitors’ lot at MARC building (north of the main entrance doors)
- Lot # 4 at MARC building (west of the west side entrance doors)
As of January 1, 2017, the rates have been increased as follows:
- Lot 1 and parking meters: $1.25 per hour with a daily maximum of $8.00
- Lot 5 (Conference/Overflow lot): $1.00 per hour with a daily maximum of $7.00