Odds Ratios vs Risk Ratios-Reply

Details

• Personal Author:
  Dowd BE ; Maciejewski ML ; Norton EC
• Description:
  In Reply: We agree with Dr Sonis that ORs have one distinct advantage over RRs when reporting the association between a binary outcome and a risk factor. If the original OR was computed for the occurrence of an outcome, then the OR for the nonoccurrence of the outcome is the inverse of the original OR. There is no such convenient transformation for RRs. He illustrates this point with a simple example and explains that this is important because for many outcomes it is arbitrary whether to report the outcome as the event occurrence or nonoccurrence. When deciding how to report the strength of association between a binary outcome and a risk factor, we want to emphasize that this property of ORs needs to be taken in context with other considerations. As discussed in our recent JAMA Guide to Statistics and Methods, ORs have important limitations. These limitations include the lack of an intuitively appealing interpretation and the fact that the magnitude of the OR depends on an arbitrary scaling factor, making direct comparison of ORs across models and studies impossible. Researchers should consider all of these issues when deciding how best to communicate results from a logistic regression model. In addition, we extend Sonis' point regarding the advantage of ORs to include risk differences, measured as the arithmetic difference in 2 risks. Unlike risk ratios, the risk difference for the nonoccurrence of the outcome attributable to a specific risk factor equals the risk difference for the occurrence of the outcome multiplied by -1. Risk differences have the additional advantage of being less sensitive than ORs to changes in the arbitrary scaling factor. Therefore, risk differences, also referred to as partial effects, are a useful way to report strength of association from a logistic regression. [Description provided by NIOSH]
• Subjects:
  Occupational Health Risk Assessment Risk Factors Safety
• Keywords:
  Mathematical Models Risk Analysis Risk Factors Statistical Analysis Statistical Quality Control
• ISSN:
  0098-7484
• Document Type:
  Text
• Funding:
  Grant
• Genre:
  Journal Article
• Place as Subject:
  Michigan ; Minnesota ; North Carolina ; OSHA Region 4 ; OSHA Region 5
• CIO:
  National Institute for Occupational Safety and Health (NIOSH)
• Topic:
  Workplace Safety & Health
• Location:
  North America
• Volume:
  320
• Issue:
  19
• NIOSHTIC Number:
  nn:20064237
• Citation:
  JAMA 2018 Nov; 320(19):2041-2042
• Contact Point Address:
  Edward C. Norton, PhD, Department of Health Management and Policy, University of Michigan, 1415 Washington Heights, Ann Arbor, MI 48109
• Email:
  ecnorton@umich.edu
• Federal Fiscal Year:
  2019
• Performing Organization:
  University of Minnesota Twin Cities
• Peer Reviewed:
  True
• Start Date:
  20050701
• Source Full Name:
  Journal of the American Medical Association
• End Date:
  20250630
• Collection(s):
  National Institute for Occupational Safety and Health
• Main Document Checksum:
  urn:sha-512:17a07ac507b1dfecf21419d20743be3674eda5cfa01844d72c444f8791256702b08ca24124c25f01fb2ce60b8ebc61c5602109dc7ca984d424a9705f51712ccb
• Download URL:
  https://stacks.cdc.gov/view/cdc/222738/cdc_222738_DS1.pdf
• File Type:
  [PDF - 66.68 KB ]