Effect Modification

Abstract

The impact of an exposure or treatment may differ across people. Effect modification occurs when the size of an association between an exposure and outcome differs according to another characteristic. A true differential impact of an exposure or treatment on the disease process is suspected when the difference in the size of an association across subgroups is large, the subgroups contain sufficient numbers of people and outcomes for comparison, there is plausibility for the observed difference, and the difference is replicated in other studies. Non-overlapping confidence intervals and a low p-value for interaction provide statistical evidence for the presence of effect modification.

The impact of an exposure or treatment may differ across people. Effect modification occurs when the size of an association between an exposure and outcome differs according to another characteristic. A true differential impact of an exposure or treatment on the disease process is suspected when the difference in the size of an association across subgroups is large, the subgroups contain sufficient numbers of people and outcomes for comparison, there is plausibility for the observed difference, and the difference is replicated in other studies. Non-overlapping confidence intervals and a low p-value for interaction provide statistical evidence for the presence of effect modification.

A pharmaceutical company sponsors a clinical trial of a new medication for treating major depressive disorder. Investigators randomly assign 5000 people who are experiencing a first major depressive episode to receive either the new medication or a selective serotonin reuptake inhibitor (SSRI), the clinical standard of care for major depression. Among the recruited participants, the mean age is 37 years, 60% are female, and 22% also meet criteria for anxiety disorder. The outcome of the study is the persistence of depressive symptoms after 6 months of treatment. Results are shown below.

Trial results stratified by the presence of baseline anxiety disorder.

	Relative risk of persistent depression symptoms comparing the new medication to SSRIs (95% confidence interval)
All participants	0.70 (0.61, 0.79)
Anxiety disorder at baseline	0.38 (0.14, 0.62)
No anxiety disorder at baseline	0.79 (0.65, 0.93)

• 72. How does the presence of anxiety disorder relate to the observed association between receipt of the new antidepressant medication and persistent depressive symptoms?

Confounder	Effect modifier
A. No	Yes
B. No	Yes
C. Yes	No
D. Yes	No
E. Cannot determine from the information that is provided

A cohort study evaluates the association of hepatitis B infection with the development of hepatic cirrhosis. Researchers identify 800 patients who have serologic evidence of active hepatitis B infection and a comparison group of 1800 healthy individuals who do not have evidence of past or present hepatitis B infection. Baseline characteristics of the study population are presented below.

	Hepatitis B positive	Hepatitis B negative
Mean age (years)	43.5	47.2
Male	68%	41%
Smoking	22%	14%
Heavy alcohol use	18%	9%
Chronic kidney disease	8%	11%

Participants are followed over 10 years to assess the occurrence of cirrhosis. The study finds that hepatitis B infection is associated with a 2.5 times higher incidence of cirrhosis:

Unadjusted relative risk 2.5; 95% confidence interval (2.3, 2.7)

Stratified study results are shown below.

Association of hepatitis B infection with cirrhosis among subgroups (relative risks and 95% confidence intervals shown)
Men	Women
2.9 (2.1, 3.7)	2.3 (1.6, 3.0)
Heavy alcohol use	No heavy alcohol use
6.8 (5.4, 8.2)	1.9 (1.6, 2.2)

• 73. Which of the following is true?

  • A. These data suggest that sex is confounding the observed association between hepatitis B infection and cirrhosis.

  • B. These data suggest that sex is modifying the observed association between hepatitis B infection and cirrhosis.

  • C. The sex-adjusted relative risk of cirrhosis associated with hepatitis B infection is between 2.3 and 2.9.

  • D. These data suggest that heavy alcohol use is confounding the observed association between hepatitis B infection and cirrhosis.

  • E. These data suggest that heavy alcohol use lies on the causal pathway between hepatitis B infection and cirrhosis.

A new vaccine designed to prevent malaria is deployed in four hypothetical countries. Disease incidences among vaccinated and non-vaccinated residents in each country are presented below.

	Incidence of malaria in non-vaccinated individuals (per 1000 at risk)	Incidence of malaria in vaccinated individuals (per 1000 at risk)
Country 1	100	40
Country 2	200	130
Country 3	300	230
Country 4	400	310

• 74. Consider the possibility that funds are available to continue the vaccination program in only one country. Deploying the new vaccine to which country would prevent the most cases of malaria?

  • A. Country 1

  • B. Country 2

  • C. Country 3

  • D. Country 4

Acute kidney injury (AKI) can result from exposure to medications that reduce blood flow to the kidneys. Two common medication classes that can decrease kidney blood flow are nonsteroidal anti-inflammatory drugs (NSAIDs) and angiotensin receptor blockers (ARBs). A case-control study evaluates the association of these medications with the incidence of AKI. Researchers identify 100 case patients who recently developed AKI and 300 control patients who have normal kidney function. The researchers then ascertain NSAID and ARB use within the previous 6 months.

Odds ratio of AKI comparing NSAID use to nonuse::

1.6 (95% confidence interval 1.25, 1.95)

Odds ratio of AKI comparing ARB use to nonuse::

2.2 (95% confidence interval 1.65, 2.75)

• 75. Which of the following procedures would be most helpful for judging whether there is a possible synergistic effect of NSAID and ARB use on the risk of AKI?

  • A. Evaluating the differential size of the odds ratios for AKI associated with NSAID and ARB use

  • B. Evaluating the 95% confidence intervals for the odds ratios of AKI associated with NSAID use and ARB use

  • C. Calculating p-values for the odds ratios of AKI associated with NSAID use and ARB use

  • D. Evaluating the association of NSAID use with AKI separately among ARB users and nonusers

Questions 76–78 refer to the article by Muzaale et al.:

• Risk of End-Stage Renal Disease Following Live Kidney Donation. JAMA. 2014;311(6):579–86.

• 76. Based on the data in the paper, how does age relate to the association between live kidney donation and the development of end-stage renal disease?

Confounder	Effect modifier
A. No	No
B. Yes	No
C. No	Yes
D. Cannot determine	Yes
E. No	Cannot determine

• 77. Which of the following is true regarding the association of live kidney donation with ESRD?

  • A. The attributable risk is greater among Black donors compared with Hispanic donors.

  • B. The relative risk is greater among Black donors compared with Hispanic donors.

  • C. The attributable risk is greater among donors who are ≥50 years old compared with donors who are <50 years old.

  • D. The relative risk is greater among donors who are ≥50 years old compared with donors who are <50 years old.

A hypothetical follow-up study obtains physical activity data for the live kidney donors and matched non-donors in the study. The association of live kidney donation with end-stage renal disease according to categories of low versus high levels of physical activity is shown below.

	Attributable risk of end-stage renal disease comparing live kidney donors to healthy non-donors
All study individuals	+26.9 per 10,000
Low physical activity	+12.3 per 10,000
High physical activity	+13.8 per 10,000

• 78. Based on these data, how do physical activity levels relate to the association between live kidney donation and end-stage renal disease?

Confounder	Effect modifier
A. No	No
B. No	Yes
C. Yes	No
D. Yes	Yes
E. Cannot determine from the information that is provided