Outcome Measures in Epidemiology

1. Summer Course:Introduction to Epidemiology August 25, 1330-1500 Outcome Measures in Epidemiology Dr. N. Birkett, Department of Epidemiology & Community Medicine, University of Ottawa

2. Session Overview • Risk (incidence) • Prevalence • Rate (incidence) • Person-years • Attack rate • Case fatality • Relationships among measures

3. Patient Scenario (1) A 60 year old, previously healthy, female research chemist recently developed shortness of breath and nosebleeds. She was pale with a rapid pulse (100 bpm). Blood tests showed a low hematocrit, elevated white blood count and reduced platelet count. A blood smear showed atypical myeloblasts. She was diagnosed with acute myelogenous leukemia (AML) confirmed by bone marrow examination.

4. Patient Scenario (2) In hospital, chemotherapy was given. Three weeks later (while still in hospital), she developed a fever (39°C) with a markedly depressed neutrophil count. There was no obvious source of infection; blood and urine cultures were taken (but no results would be available for 2 days). She was treated with a cocktail of broad-spectrum antibiotics . Cultures confirmed a Staphylococcus aureus infection.

5. Patient Scenario (3) • Potential research questions • Why did she get leukemia? • Does exposure to chemicals at work increase risk of leukemia? • Are new safety procedures needed? • What is the best treatment? • Does chemotherapy work? • What is the trade-off between quantity and quality of life? • How important are infections in leukemia patients? • What can we do to prevent infections? • How can we treat them? • What is the best treatment?

6. Patient Scenario (4) • Risk of AML is 10-times higher in persons with long-term exposure to benzene. • Prognosis is poor: 20% five year survival • Infection is a common problem in patients with cancer, often related to chemotherapy suppression of bone marrow. • Wide range of bacterial pathogens. • 50% of patients with fever won’t have a source of infection identified. • Empirical treatment based on epidemiological evidence of ‘common’ organisms. • Patients developing fever are routinely treated with broad spectrum antibiotics

7. Measures (1) • Patient scenario used several ‘measures’ • Five-year survival is only 1 in 5 (20%). • Benzene exposure increases risk by 10 times • 50% of cancer patients with fever have no cause • Empirical antibiotic treatment selected based on distribution of infections found in similar patients.

8. Measures (2) • Core measures of disease occurrence • Risk (cumulative incidence; incidence proportion) • Prevalence • Incidence rate (incidence density)

9. Risk (1) • Definition • The proportion of unaffected individuals who, on average, will develop the disease (or outcome) of interest over a specific period of time. • Simplest method of estimation is to follow a group of unaffected people for a fixed period of time and count the number who develop the outcome. Then:

10. Risk (2) • Risk is an incidence measure. • Risk lies between 0 and 1 • It is a probability and has no units. • It is sometimes expressed as a percent. • Consider this simple example from the book (figure 2.1): • 6 subjects were studied for at least two years between 1995 and 2004 • aside: the book has the wrong x-axis labels in Fig 2-1. • What is risk of developing disease in first two years of follow-up?

11. Risk (3)

12. Risk (4) • For risk, we don’t really care when (in calendar time) the people were diagnosed. • All we are concerned about is if they developed the disease and when, after they entered the study, they developed it • It is standard to change the time scale so that each person entered the study at time ‘0’. • In some cases, you won’t do this, but we won’t worry about those 

13. Risk (5)

14. Risk (6) • There are 6 people in the cohort. They were all followed for at least two years. • In those two years, one person (A) developed the disease (outcome). • Therefore, the risk of developing the outcome in 2 years is:

15. Risk (7) • What if want the risk of developing disease in 3 years? • Problem is that two people (B and E) were only followed for two years. • B was ‘lost’ before developing the disease. • E ‘ran out’ of study time. • For neither of these people do we know if they would have developed disease if they had been followed for three years. • Two choices (we’ll return to these later): • Compute ‘incidence rate’ and convert to risk. • Survival analysis methods

16. Prevalence (1) • Definition • The proportion of individuals in a population at a point in time who have the condition of interest • There is NO time dimension involved with point prevalence

17. Prevalence (2) • Some confusion exists with names for ‘prevalence’. • Technically, we are defining point prevalence. • Period prevalence is a related concept but is not commonly used. • Sometimes called ‘prevalence rate’. • Some authors (e.g. K. Rothman) use ‘prevalence’ to refer to the ‘number’ of cases. • Proportion of people with the event is called ‘prevalence proportion’ instead of ‘prevalence rate’. • In our example, what is disease prevalence in 2001?

18. Prevalence (3)

19. Prevalence (4) • At the start of 2001, there 4 people being actively studied. They make up the population. • Of these 4 people, one person (A) has the disease at the start of 2001. • Therefore, the prevalence of the disease is:

20. Prevalence & incidence (1) • What general conditions produce a high prevalence? • If there are lots of people getting ill (incidence), one would expect more to be ill on a particular day (prevalence) • If people stay ill for a longer time period (duration), more should be ill on any given day (compare cancer to a cold)

21. Prevalence & Incidence (2)

22. Prevalence & incidence (3) • As long as conditions are ‘stable’, we have this relationship:

23. Incidence rate (1) This is the hardest one to get a good grasp of. • The Basic Concept • Measures the rapidity with which newly diagnosed cases of the disease (or outcome) develop. • This is similar to the speed of driving a car. • Incidence rate is particularly useful when people are followed for different lengths of time. • Lost to follow-up • Die before developing outcome • Study runs out of money.

24. Incidence rate (2) • Several methods of estimating incidence rate exist. • For those with a strong statistical background, Incidence Rate is essentially the same as the hazard function (survival analysis) or the force of mortality (demography). • Simplest method of estimation is to follow a group of unaffected people for a period of time, count the number who develop the outcome and track the ‘person-time at risk’. Then:

25. Incidence rate (3) • What is person-time? • The total amount of time which people are under observation. • Follow 5 people for 2 years • There are 10 person-years of follow-up (PYs). • Follow 2 people for 5 years • You also have 10 Person years of follow-up (PYs).

26. Incidence rate (4) • What is person-time? (cont.) • Computing PT: • Add up the follow-up time for every subject in the cohort

27. Incidence rate (6)

28. Incidence rate (7) • What is incidence rate of developing the disease outcome in our mini-example? • # new cases = 2 (subjects A and C). • Person time:

29. Incidence rate (7) • # new cases = 2 • PT or PY’s = 22 • Therefore, the rate of developing the outcome is: • Assumes that the rate is constant over the follow-up period.

30. Incidence rate (4) • What is person-time? (cont.) • Computing PT: • Add up the follow-up time for every subject in the cohort • Use an approximation: • PT = population size * time of follow-up • Most useful when working with large population groups (e.g. the entire Canadian population). • Can get more complicated if you want to compute age or sex specific rates. • Need to accumulate person time within each age/sex category. • Now all done by computer but you need to know about the Lexis diagram.

31. Incidence rate (8) 52 51 50 49 1998 1999 2000 2001

32. Relationship of risk and rate (1) • Consider 1000 people, followed for 1 year • 10 get the disease. • How do the incidence risk (over 1 year) and incidence rate relate to each other? Risk

33. Relationship of risk and rate (2) Incidence rate • We need person-time. • People who get disease are only ‘at risk’ before they get ill. • Time of disease onset is not given to us • Assume it occurs one half way through year • Actuarial assumption • 990 people did not get ill  1 PY each • 10 people got ill  0.5 PY each

34. Relationship of risk and rate (3) Incidence rate

35. Relationship of risk and rate (4) • Incidence risk = 0.01 • Incidence rate = 0.01005 cases/person-year • In this simple situation, we have: rate ≈ risk • The same to within 1%

36. Relationship of risk and rate (5) • More generally, we have: • This is true as long as the risk is low • say < 0.05 or 5%. • A more accurate conversion is: • These formulae all assume that the rate is constant over time • Very unlikely true • But commonly used in epidemiology.

37. Attack rate (1) • A traditional measure used in studies of disease outbreaks, mainly due to infectious disease agents. • It is the probability that a person at risk of the disease will develop disease as a result of the outbreak. • It is incorrectly called a ‘rate’. • It is actually a ‘risk’. • Needs a time referent to be complete. • It is very similar to incidence risk

38. Attack rate (2) • Between 10:00 pm, January 17 and 8:00 pm, January 18: • 47 college students living in residence developed an acute gastrointestinal disease. • 1,164 students lived in residence and were considered ‘at risk’ for the GI disease. • The attack rate is:

39. Case Fatality Rate (1) • The probability that a person with a disease will die from the disease. • Most commonly applied to acute infections • Competing mortality not an issue with acute illness • Provides a measure of the severity of the disease. • CFR(rabies) = 1.0 • CFR(‘cold’) ≈ 0.000001

40. Case Fatality Rate (2)

41. Case Fatality Rate (3) • Mortality rate depends on CFR and incidence • If the CFR is high, then the mortality rate will be higher for a given incidence. • If there are ‘lots’ of cases (high incidence), then the mortality rate will tend to be higher for a given CFR.

42. Survival (1) • The complement to Cumulative Incidence is SURVIVAL • the probability that a person with a disease will still be alive after a certain number of years. • Survival is usually denoted as S(t). • Simplest way to estimate survival:

43. Survival (2)

44. Survival (3) • Let’s compute survival for 2 years and 5 years in this small sample. • 2 years • 6 people are ‘at risk’ in study • Everyone was followed for at least two years. • One person (C) died before two years. • S(2) = 5/6 = 0.83 • Fairly direct 

45. Survival (4) • 5 years • 6 people are ‘at risk’ in study • 2 people died before five years (C and F). • 2 people were still alive after five years (A & D) • What do we do about people B & E? • Assume they died: • S(5) = 2/6 = 0.33 • Assume they lived 5 years: • S(5) = 4/6 = 0.66 • A big difference

46. Survival (5) • Survival analysis is designed to handle censored data • Life tables • Actuarial method • Kaplan-Meier method • Technical details are beyond this course. • We’ll cover the basic ideas.

47. Survival (6) 10 dead dead 15 100 90 alive 75 alive |____________________|___________________| 0 1 2 Pr(survive 1 year) = 90/100 = 0.9 Pr(survive 2nd year|survive 1st year) = 75/90 = 0.83 Pr(survive both years) = 0.9*0.83 = 0.75

48. Survival (7) • Can be extended to handle people who are lost to follow-up, etc. (censored cases) • Key features • Break follow-up into intervals • Usually 1 year for actuarial life tables • Estimate the ‘at risk’ population in each interval by assuming that 50% of censored group were at risk. • The actuarial assumption • Compute probability of surviving each year conditional on surviving previous years • Multiply to give overall survival

49. Summary • Key measures • Prevalence • Incidence risk • Incidence rate • Person-time • Survival measures • Needed to handle issues such as: lost subjects, incomplete follow-up and competing mortality.