I performed a multivariate logistic regression to estimate the role of some baseline variables (e.g. age, sex, etiology etc.) on a long-term outcome (good, bad). I only have one patient group and no patient underwent any treatment.
Can I calculate absolute risk reduction (ARR) or number needed to treat (NNT) in this case?
Hi,
No, it is not possible. You need at least one group exposed to any factor of interest and one group not exposed to calculate absolute risk reduction.
Best regards.
See these papers
https://academic.oup.com/ndt/article/32/suppl_2/ii13/3056571
Article Tips for learners of evidence-based medicine: 1. Relative ri...
You asked is it possible, and is it sensible. I think the answers are 'yes' and 'yes if you can help people interpret it'. Is a lifestyle modification a 'treatment'? In a group of tobacco smokers, we could define quitting smoking as a 'treatment' vs. continuing to smoke as the control, and calculate ARR to get the answer 'what is the difference in the proportion of ex-smokers and smokers who get Disease Y in 10 years?' Taking the reciprocal 1/ARR, you get NNT, which is 'how many smokers would have to quit smoking to prevent one case of Disease Y in 10 years?'
Conceptually, the interpretations of these effect statistics are the same whether the 'treatment' is a pill to block nicotine receptors or a New Year's resolution to quit smoking. This of course assumes that the observed differences in event proportions are causal (not confounded by self-selection if it's a cohort study, or bias due to lack of concealed allocation if it's an RCT), or some other biasing mechanism such as unequal followup time. NNT is supposed to be calculated on cumulative incidence and can be misleading if simple proportionate outcomes are analyzed but followup times differ between exposure groups. (Ref: Suisa S, The Number Needed to Treat: 25 Years of Trials and Tribulations in Clinical Research, Rambam Maimonides Med J www.rmmj.org.il 1 July 2015 Volume 6 Issue 3 e0033)
NNT arose from clinical epidemiology and if properly calculated is a simple and easily understood measure of effect (usually of 'good' exposures) for patients, physicians and policy makers. In population epidemiology and public health you'd be more likely to encounter attributable fraction (population) and attributable fraction (exposed), usually for 'bad' exposures. The Number Needed to Harm is the clin epi version for bad exposures. Number needed to screen, and to vaccinate, are the same basic effect measure applied to other public health interventions.
All of the above require the researcher and the consumer to understand the valid interpretations of absolute, difference (by subtraction) and relative (by division) statistics:
Two absolute rates: in Denmark, where most baby boys are not circumcised, the incidence of invasive penile cancer is about 10 per million. In Israel where most baby boys are circumcised the incidence of IPC = 1 per million. Interpretation: IPC is very rare in Denmark, and very very rare in Israel.
The ratio measure (RR) is 10 / 1 = 10. Interpretation: the risk of IPC is 10 times higher in uncircumcised boys.
The absolute difference measure is 10 - 1 = 9 per million. Interpretation: Assuming causation, there would be nine fewer cases of IPC per million if all baby boys were circumcised.
ARR = (10/1,000,000) - (1/1,000,000) = .00001 - .000001 = .000009
NNT = 1/.000009 = 111111.1 Interpretation: about 111,112 boys would have to be circumcised to prevent one case of IPC.
- Badges
- Science topic
- Similar topics
- Medicine
- Disease
- Infectious Diseases