Simon Thornley Field epidemiology: effect measures, and a cohort study of a food borne outbreak
Overview Introduce how epidemiology can be used to help inform what caused a food borne outbreak Re-inforce ideas of cohort study design and analysis Consider how other evidence can inform the results of epidemiological study.
Basic study design
Participants Outcomes Exposure
Simple Outbreak 6 October people reported having diarrhoea to North Shore City Council following ‘prize giving’ at North Shore Event Centre Prize giving was the final event of a week long, international soccer tournament of the South African/Indian, diaspora
First steps: what to do? Outbreak group Track down participants Track down food Microbiological samples Assess risk Immediate action required? Media/ risk communication?
Hypothesis Was this outbreak caused by food served at function? Which food? Epidemiological analysis one tool. Clinical and microbiological analysis provide complementary
The suspects... Lamb biryani Vege biryani Dahl
The premises
Consider raw food
RETROSPECTIVE COHORT Buffers’ dance Dal/ biryaniDiarrhoea?
Steps in assessing data
Symptom onset (outbreak curve) :30:00 p.m. 1:30:00 a.m.3:30:00 a.m.5:30:00 a.m.7:30:00 a.m. 9:30:00 a.m. 11:30:00 a.m. 1:30:00 p.m.3:30:00 p.m.5:30:00 p.m.7:30:00 p.m. 9:30:00 p.m. 11:30:00 p.m. 1:30:00 a.m.3:30:00 a.m. Time of diarrhoea onset frequency
Common epidemic curves
Symptoms? SymptomNumberProportion (%) Diarrhoea5192.7% Bloody diarrhoea 35.5% Abdominal cramps % Nausea1323.6% Fever35.5% Vomiting59.1% Headache1527.3%
Incubation period IllnessIncubation, hours (n=51)* Duration, hours (n=43)* Median12’30”21’00” Mean11’36”21’44” Range3’30”-33’00”2’00”-50’00”
What are the facts? Facts vs artifacts? Bias: selection, information Selection: was the sample representative of the total group? Would New Zealand cases be more/less likely to be sick? Case definition?
Diagrams helpful
“Frequentist view” Ground hog day Imagine the event repeated over and over! Assume random probability
Error distribution Assume outcome is like flipping a biased coin (probability theory) Consider long run probability associated with biased coin (prob =attack rate) Bias = attack rate in exposed or unexposed groups
Crude associations (univariate or crude) Eaten? YesEaten? No FoodIllWellAttack rate (%) IllWellAttack rate (%) OddsP Ratio (95% CI) Dahl (1.3, 10.0)0.02 Lamb Biryani (0.6, 38.6)0.15 Vegetable Biryani (0.1, 0.98)0.04 Salad (0.5, 2.2)0.61
In pictures (dal) Dahl eaten Diarrhoea If unrelated; chance of diarrhoea after eating dahl =chance of diarrhoea (regardless of meal selection) Actual No effect
In pictures (dal) Diarrhoea Actual Dahl exposure Odds Ratio =0.77/0.22 =3.5
In pictures (dahl) Diarrhoea No effect Dahl exposure Odds Ratio =0.61/0.65 ≈1 (no effect)
In pictures (lamb) Lamb eaten Diarrhoea Actual No effect
In pictures (vege) Diarrhoea Actual No effect
Just consider first exposure! Is dal the likely culprit? What information does the p-value give you? What information does the confidence interval convey? What is the null hypothesis?
Risk of disease in dal eaters
Odds of disease in dal eaters
Risk of disease in non-dal eaters
Odds of disease in non dal eaters
Distribution of risk ratio Upper bound of risk ratio is reciprocal of risk in unexposed (28/5 or 5.6)
Distribution of odds ratio
What if we’d managed to get 10x participants?
What do these results mean? What food caused the outbreak? What does the 95% confidence interval mean? Could these results be explained by chance? Are the populations comparable? Was some other exposure accounting for this difference?
How do we know differences are real (not due to chance)? We don’t – but Statistics helps us put a number on the uncertainty !
Confounding: Are the populations similar? VariableCases (n=51)Non-cases (n=83) P-value Gender 0.5 Male, n (%)24 (47%)33 (40%)(chi-square) Age (years) Range Mean4541 (unpaired t- test)
Multivariable Results ExposureCrude Odds ratio (95% CI) Adjusted Odds ratio (95% CI) Dal 3.5 (1.3 to 10.0) 3.7 (1.3 to 10.9) Vegetable Biryani 0.4 (0.1 to 1.0)0.5 (0.2 to 1.6) Lamb Biryani 4.6 (0.6 to 38.6)0.8 (0.1 to 5.3)
Microbiology Stool (2/18 received) Clostridium perfringens (1.0 x 10 4 CFU/g) Bacillus cereus No toxin, no norovirus Food Lentils ok Spices Turmeric Bacillus cereus (1.0 x104 CFU/g, with faecal coliforms (4.0 MPN/g)
Reject null hypothesis dal likely to cause outbreak Relationship between illness and dal confounded by lamb biryani
Putting it all together…. Turmeric seeded lentil soup with Bacillus cereus Time and temperature abuse – endotoxin associated diarrhoeal syndrome. Vegetable biryani protective, few consumed, more likely to be heated in bain-marie. Likely temperature labile toxin.
Room set up Lamb biryani and dal Vege biryani and dal
Don’t forget the Public Health Action Is routinely contaminated turmeric ok? What should we do with the cook? Legal action justified?
Summary Epidemiology- distribution and determinants of health Study design – take into account factors such as cost, speed, causation, ethical issues Statistics P-value yes/no cf. 95% confidence interval Put results together with other analysis Unexpected results!