In the 19th century health was transformed by clear, clean water. In the 21st century health will be transformed by clean, clear knowledge. Sir Muir Gray 70-80% of Doctors Do Not Understand Health Statistics: What Can Be Done? Gerd Gigerenzer Max Planck Institute for Human Development Berlin Improving Health Literacy Problem: Most doctors and patients do not understand health statistics. Causes: - Failure of medical schools to teach statistical thinking. - Biased reporting. Solution: 1. Teach risk literacy in medical school. 2. Implement transparent risk communication. Gigerenzer & Muir Gray (Eds.) 2011. Better doctors, better patients, better decisions. MIT Press. I. Most doctors do not understand health statistics Gigerenzer & Muir Gray (Eds.) 2011. Better doctors, better patients, better decisions. MIT Press. "I had prostate cancer, five, six years ago. My chances of surviving prostate cancer and thank God I was cured of it, in the United States, 82 percent. My chances of surviving prostate cancer in England, only 44 percent under socialized medicine.” Rudy Giuliani, New Hampshire radio advertisement, October 2007 Lead Time Bias Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz, & Woloshin 2007. Psychological Science in the Public Interest. Overdiagnosis Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz, & Woloshin 2007. Psychological Science in the Public Interest. Do German Physicians Understand 5-Year Survival Rates? 65 physicians (internal medicine) When the (same) information was framed as Survival rates: 79% judged screening as effective Mortality rates: 5% judged screening as effective Lead-time-bias? Overdiagnosis? 2 out of 65 knew 0 out of 65 knew Wegwarth, Gaissmaier & Gigerenzer, Medical Decision Making, 2011 Do U.S. Physicians Understand 5-Year Survival Rates? 412 primary-care physicians (national sample) Survival rates: Mortality rates: 83% judged mortality benefit as large 28% judged mortality benefit as large Which proves that a cancer screening test “saves lives”? 1. Screen-detected cancers have better 5-year survival. 76% 2. More cancers are detected in screened populations. 47% 3. Mortality rates are lower among screened persons. 81% Wegwarth, Schwartz, Woloshin, Gaissmaier & Gigerenzer, Annals of Internal Medicine, 2012. Deception by Medical Institutions One of the most prestigious cancer centers in the US: M. D. Anderson Confusion about progress against cancer. Unwarranted enthusiasm for medical center. Deception by Pink Ribbon Organizations: Susan G. Komen Early detection saves lives. The 5-year survival rate for breast cancer when caught early in 98%. When it’s not? 23%. Deception by Pink Ribbon Organizations: Susan G. Komen Early detection saves lives. The 5-year survival rate for breast cancer when caught early in 98%. When it’s not? 23%. II. Helping Doctors and Patients Make Sense of Health Statistics: Fact Boxes Schwartz, Woloshin & Welch 2009. Annals of Internal Medicine Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz, & Woloshin 2007. Psychological Science in the Public Interest. Fact Box Breast cancer screening with mammography: per 1,000 women 50+ No Screening Screening over 10 years Benefits? Breast cancer mortality 5 4 Total cancer mortality 21 21 False positives with biopsies – 50 – 200 Unnecessary treatments (e.g. lumpectomy) – 2 - 10 Harms? Gøtzsche PC & Jørgensen KJ 2013. Cochrane Database of Systematic Reviews Woloshin S & Schwarz LM 2009. Journal of the National Cancer Institute 101(17) www.harding-center.mpg.de THE ART OF RISK COMMUNICATION HOW BREAST CANCER PAMPHLETS MISLEAD WOMEN “Breast screening has been shown to reduce the risk of dying from breast cancer by around 35%.” The Welsh NHS leaflet Breast Screening “Screening mammograms . . . reduce the chance of dying from breast cancer by approximately 33%.” New Zealand Breast Cancer Foundation “Screening saves about 1 life from breast cancer for every 200 women who are screened.” Gigerenzer 2014 British Medical Journal NHS Leaflet for England What Does a 25% Reduction of Breast Cancer Mortality Mean? 15 Gynecologists, Department of Gynecology Luzern Schüssler 2005, Frauenheilkunde aktuell Prostate Cancer Early Detection by PSA screening and digital-rectal examination. Numbers are for men aged 50 years or older, not participating vs. participating in screening for 10 years. 1,000 men without screening: 1,000 men with screening: P PP P P PP P P PP P P PP P XXXXXXXXXXXXXXXXXXXX P X Men dying from prostate cancer: 8 8 Men dying from any cause: 200 200 Men that were diagnosed and treated for prostate cancer unnecessarily: – 20 Men without cancer that got a false alarm and a biopsy: – 180 Men that are unharmed and alive: 800 600 Source: Djulbegovic, Beyth, Neuberger et al. 2010. British Medical Journal. “I never dreamed that my discovery four decades ago would lead to such a profit-driven public health disaster. The medical community must confront reality and stop the inappropriate use of PSA screening. Doing so would save billions of dollars and rescue millions of men from unnecessary, debilitating treatments.” Richard Ablin, 2010 discoverer of PSA (prostate specific antigen) Harmful Overtreatment In 2014, German gynecologists recommended ultrasound screening to 3 million women. About 2 million ultrasounds were performed. Women paid 75 million Euro from their own pockets. About 35,000 healthy women lost their ovarian. Health insurances paid > 100 million Euros for surgery and treatment of consequences. www.aok.de/faktenboxen III. Helping Doctors and Patients Understand Screening Tests: Natural Frequencies The Illusion of Certainty What Does a Positive Test Mean? A positive result means antibodies to HIV were found in your blood. This means you have HIV infection. You are infected for life and can spread HIV to others. You will most likely develop AIDS, but no one can know when you will get sick. Within 10 years after infection, about half of untreated people have developed AIDS. Illinois Department of Public Health A letter from California Conditional Probabilities Natural Frequencies 10,000 clients 9,999 no HIV 1 HIV 1 positive p(HIV) = 0.01% p(positive|HIV) = 99.9% p(negative|no HIV) = 99.99% 0 negative 1 positive 9,998 negative Conditional Probabilities Natural Frequencies 10,000 clients 9,999 no HIV 1 HIV 1 positive p(HIV) = 0.01% p(positive|HIV) = 99.9% p(negative|no HIV) = 99.99% 0 negative 1 positive 9,998 negative p(HIV|positive) = .0001x.999 .0001x.999 + .9999x.0001 Conditional Probabilities Natural Frequencies 10,000 clients 9,999 no HIV 1 HIV 1 positive p(HIV) = 0.01% p(positive|HIV) = 99.9% p(negative|no HIV) = 99.99% 0 negative 1 positive 9,998 negative p(HIV|positive) p(HIV|positive) = 1 1+1 = .0001x.999 .0001x.999 + .9999x.0001 Twenty AIDS Counseling Centers Client: "If one is not infected with HIV, is it possible to have a positive test result? “ Counselor: 1 2 3 4 5 6 7 8 9 10 "No, certainly not“ "Absolutely impossible“ "With absolute certainty, no“ "No, absolutely not“ "Never“ "Absolutely impossible“ "Absolutely impossible“ "With absolute certainty, no“ "The test is absolutely certain“ "No, only in France, not here" 11 12 13 14 15 16 17 18 19 20 "False positives never happen“ "With absolute certainty, no" "With absolute certainty, no“ "Definitely not" … "extremely rare“ "Absolutely not" … "99.7% specificity“ "Absolutely not" … "99.9% specificity“ “More than 99% specificity“ “More than 99.9% specificity“ "99.9% specificity“ “No comment" Gigerenzer, Hoffrage & Ebert (1998), AIDS CARE. 160 Gynecologists Estimate the Probability of Breast Cancer Given a Positive Screening Mammogram 90 Estimates in % 81 10 1 Before training Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz, & Woloshin 2007. Psychological Science in the Public Interest. Teaching Gynecologists to Use Natural Frequencies Fosters Insight 90 Estimates in % 81 10 1 Before training After training Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz, & Woloshin 2007. Psychological Science in the Public Interest. Natural Frequencies Conditional Probabilities 1000 women 1 woman 10 cancer 9 positive 990 no cancer 1 negative 89 positive 901 negative p(cancer|positive) = 9 9 + 89 .01 cancer .90 positive .99 no cancer .10 negative .09 positive p(cancer|positive) = .01 x .90 .01 x .90 + .99 x .09 Gigerenzer & Hoffrage Psychological Review 1995, 1999 .91 negative Fourth-Graders Can Make Bayesian Inferences Using Natural Frequencies • • • Out of every 20 pupils in the Ravenclaw School of Magic, 5 have a wand. Of these 5 pupils, 4 also wear a magical hat. Of the other 15 without wands, 12 also wear a magical hat. Imagine a group of pupils who wear a magical hat. How many of those also have a wand? Gigerenzer 2014. Risk Savvy. Viking Impact on Medical Practice Health Information Policy: •Misleading statistics have been removed from (almost) all cancer screening pamphlets in Germany. • The first fact boxes published in Austria by the Tyrolean Society for General Medicine. Medical Education: • Members of the Harding Center have trained 3,000 physicians in CME. • Medical schools have begun teaching methods of risk communication (e.g. Charité, Giessen, Mainz). Communication Tools: •Using “natural frequencies” (Gigerenzer & Hoffrage, Psychological Review 1995, 1999) has become a standard method in evidence-based medicine, recommended by the Cochrane Collaboration, the International Patient Decision Aid Standards Collaboration, and the Medicine and Healthcare Products Regulatory Agency. Problem: Most doctors and patients do not understand health statistics. Causes: Failure of medical schools. - Biased reporting. - Lack of education of the public. Solutions: 1. Teach risk literacy in medical school. 2. Implement transparent risk communication. 3. Teach health literacy in school, starting early. MIT Press Problem: Most doctors and patients do not understand health statistics. Causes: Failure of medical schools. - Biased reporting. - Lack of education of the public. Solutions: 1. Teach risk literacy in medical school. 2. Implement transparent risk communication. 3. Teach health literacy in school, starting early. RISK VS UNCERTAINTY RISK: How should we make decisions when all relevant alternatives, consequences, and probabilities are known? STATISTICS UNCERTAINTY: How should we make decisions when NOT all alternatives, consequences, and probabilities are known? HEURISTICS Gigerenzer, Todd & ABC Research Group 1999. Simple Heuristics That Make Us Smart. OUP The heart disease predictive instrument (HDPI) Chest Pain = Chief Complaint EKG (ST, T wave ∆'s) History No MI& No NTG MI or NTG MI and NTG ST&T Ø 19% 27% 37% ST 35% 46% 58% T 42% 53% 65% ST 54% 64% 75% ST&T 62% 73% 80% ST&T 78% 85% 90% ST 21% 29% 40% T 26% 36% 47% ST 36% 48% 59% ST&T 45% 56% 67% ST&T 64% 74% 82% ST 9% 14% 20% T 12% 17% 25% ST 17% 25% 35% ST&T 23% 32% 43% ST&T 39% 51% 62% Chest Pain, NOT Chief Complaint EKG (ST, T wave ∆'s) History No MI& No NTG MI or NTG MI and NTG ST&T Ø 10% 16% 22% No Chest Pain EKG (ST, T wave ∆'s) History No MI& No NTG MI or NTG MI and NTG ST&T Ø 4% 6% 10% See reverse for definitions and instructions Fast-and-Frugal Tree for Coronary Care Unit Allocation ST segment changes? no yes Chief complaint of chest pain? no yes Regular nursing bed Any one other factor? (NTG, MI,ST,ST,T) no Regular nursing bed Green & Mehr (1997) Coronary Care Unit yes Coronary Care Unit Emergency Room Decisions: Admit to the Coronary Care Unit? 1 Sensitivity Proportion correctly assigned .9 .8 .7 .6 Physicians .5 Heart Disease Predictive Instrument .4 Fast and Frugal Tree .3 .2 .1 .0 .0 .1 .2 .3 .4 .5 .6 .7 .8 .9 False positive rate Proportion of patients incorrectly assigned 1 Fast-and-Frugal Tree for Assessing Bank Vulnerability Leverage ratio < 4.1%? Yes No Market-based capital ratio < 16.8% No Yes Loan to deposit ratio > 1.4? No GRE EN FLA G RED FLAG Yes RED FLAG Leverage ratio = RED FLAG Market-based capital ratio = Tier 1 capital Total assets Market capitalization Risk-weighted assets Loan to deposit ratio = * 100 * 100 Retail loans Retail deposits Aikman, Galesic, Gigerenzer, Kapadia, Katsikopoulos, Kothiyal, Murphy & Neumann 2014, Bank of England Financial Stability Paper, No. 28 A Signal-Detection-Analysis of Fast-and-Frugal Trees Luan, Schooler & Gigerenzer Psychological Review 2011 A Signal-Detection-Analysis of Fast-and-Frugal Trees Bank failure CCU allocation Checkpoint HIV Luan, Schooler & Gigerenzer Psychological Review 2011 Problem: Most doctors and patients do not understand health statistics. Causes: Not irrationality, but - Failure of medical schools. - Biased reporting. - Lack of education of the public. Solutions: 1. Teach risk literacy in medical school. 2. Implement transparent risk communication. 3. Teach health literacy in school, starting early. MIT Press Problem: Most doctors and patients do not understand health statistics. Causes: Not irrationality, but - Failure of medical schools. - Biased reporting. - Lack of education of the public. Solutions: 1. Teach risk literacy in medical school. 2. Implement transparent risk communication. 3. Teach health literacy in school, starting early. Doctors and Patients Should Have the Right to Clear Information - Ethics committees should recognize that misleading the public is a key moral issue. - Fact Boxes should be made available . - Free access to the Cochrane Library www.thecochranelibrary.com - Honest health pamphlets. Gigerenzer & Muir Gray (eds.) 2011. Better doctors, better patients, better decisions. MIT Press. If you open a book on rat DM, people lack rationality, commit long lists of CI, which are compared to stable Visual illusions. OC, base rate neglect … All of this points to the conclusion that Homo Sapiens is a misnomer. Are people stupid? Economist: “Human beings are fallible; lazy, stupid, greedy and weak.” Influential scholars argued that we are hopeless when dealing with risk: “our species is uniformly probability-blind” … Collect the experts, close the doors, and tell people what to do. Or find out how to ‘nudge’ them into proper behavior. This fatalistic message is not what you will hear this evening. The problem is not individual stupidity but the phenomenon of a risk-illiterate society. -Everyone can learn to deal with risk and uncertainty. -Experts are part of the problem rather than the solution. Gazzaniga, President’s Coucil on Bioethics (endless discussions) NCI: President appoints the members of the NCI advisory committee directly. Estimates in % Colorectal Cancer correct estimate conditional probabilities Gigerenzer (2002), Reckoning with risk. Penguin. (US version: Calculated risks. Simon & Schuster.) Estimates in % Colorectal Cancer correct estimate conditional probabilities natural frequencies Gigerenzer (2002), Reckoning with risk. Penguin. (US version: Calculated risks. Simon & Schuster.)