Number needed to harm examples

Steve Simon


I was trying to track down a reference on the NNH (number needed to harm) for antibiotic use, but noticed instead the large number of good examples of NNH calculations in journal articles with full text on the web.

The adverse neuro-developmental effects of postnatal steroids in the preterm infant: a systematic review of RCTs. K. J. Barrington. BMC Pediatr 2001: 1(1); 1. [Medline] [Abstract] [Full text] [PDF]

Postnatal steroid treatment is a bad idea because it increases the chances of cerebral palsy and neuro-developmental impairment. The numbers needed to harm are 7 and 11, respectively.

Systematic review and meta-analysis of early postnatal dexamethasone for prevention of chronic lung disease. T. Bhuta, A. Ohlsson. Arch Dis Child Fetal Neonatal Ed 1998: 79(1); F26-33. [Medline] [Abstract] [Full text] [PDF]

Dexamethasone is a treatment for chronic lung disease in very low birthweight infants. Three different timings of medication were studied, and the one that showed the greatest effect was when dexamethasone was started within 7 to 14 days of birth. The NNT is 8 for mortality, meaning that you would have to treat 8 infants on average with dexamethasone to prevent one death. The NNT for prevention of chronic lung disease (CLD) at 28 days was 6, meaning that you would have to treat 6 infants on average to prevent one case of CLD. There was a significant increase in the risk of hypertension in this group, with an NNH of 65. You would have to treat 65 patients, on average, in order to see one additional case of hypertension. The authors mention that this hypertension was temporary, so this seems like a side effect well worth tolerating.

Finasteride in the treatment of clinical benign prostatic hyperplasia: a systematic review of randomised trials. J. E. Edwards, R. A. Moore. BMC Urol 2002: 2(1); 14. [Medline] [Abstract] [Full text] [PDF]

Finasteride has a significant effect on prostate volume, maximum urinary flow rate, and symptom scores. These are continuous measures and thus are presented as changes in mean scores rather than as NNTs. The prostate volume decrease from 43.7 cubic cm to 32.7 in the finasteride group, but only from 44.8 to 43.0 in the control group. The urinary flow rates increased from 11.2mL/s to 12.5 after 24 months in the finasteride group, while they increased only from 10.5 to 11.3 in the placebo group. Thus the finasteride group had a 0.5 mL/s better improvement, on average. At 12 months, the average symptom score was 3.7 points lower in the finasteride group compared to a 2.3 point decline with placebo. The symptom score, developed by the American Urological Association ranges between 0 and 35.

I’m not a urologist, but other than prostate volume, these changes look small and of questionable clinical relevance.

The authors did show an NNT of 49 and 26 for avoiding acute urinary retention at 24 months and at 48 months. The NNTs for avoiding prostate surgeries were 31 and 18 for 24 and 48 months, respectively. Both acute urinary retention and prostate surgery seem like rather extreme events, so these NNTs would still be attractive.

There were several NNH calculations, as finasteride had an increased risk for various side effects. The NNHs were 14 (sexual dysfunction), 47 (decrease libido), 24 (impotence), and 55 (ejaculation disorder).

Here’s where some value judgments come into play. How important is it to decrease prostate volume and to avoid acute urinary retention and surgery? Is the risk of various types of sexual dysfunction worth the benefits? The benefits do seem to outweigh the risks to me, but different people may come to different conclusions.

As a side note, this study avoided heterogeneity tests and funnel plots because “they lack the power to reliably detect statistical heterogeneity or publication bias” and use sensitivity analyses instead. I will try to write up something about this issue soon.

Sildenafil (Viagra) for male erectile dysfunction: a meta-analysis of clinical trial reports. R. A. Moore, J. E. Edwards, H. J. McQuay. BMC Urol 2002: 2(1); 6. [Medline] [Abstract] [Full text] [PDF]

Sildenafil seems to work well. In the sildenafil group, 49% of the men has a good outcome (meaning at least 60% of the attempts at sexual intercourse were successful). In the placebo group only 11% had good outcomes. This leads to an NNT of 2.7.

Inhaled corticosteroid doses in asthma: an evidence-based approach. H. Powell, P. G. Gibson. Med J Aust 2003: 178(5); 223-5. [Medline] [Full text] [PDF]

Single-dose ketorolac and pethidine in acute postoperative pain: systematic review with meta-analysis. L. A. Smith, D. Carroll, J. E. Edwards, R. A. Moore, H. J. McQuay. Br J Anaesth 2000: 84(1); 48-58. [Medline] [Abstract] [PDF]

There were also two interesting methodology articles which are worth looking at.

Linking evidence-based medicine therapeutic summary measures to clinical decision analysis. B. Djulbegovic, I. Hozo, G. H. Lyman. MedGenMed 2000: 2(1); E6. [Medline] [Full text]

Reporting risks and benefits of therapy by use of the concepts of unqualified success and unmitigated failure: applications to highly cited trials in cardiovascular medicine. G. B. Mancini, M. Schulzer. Circulation 1999: 99(3); 377-83. [Medline] [Abstract] [Full text] [PDF]

You can find an earlier version of this page on my original website.