From: Number needed to treat (NNT) in clinical literature: an appraisal

Study | Variable | Baseline risk | Time horizon | Confidence interval | Methodology used to compute NNT defined in methods section | Method used to compute NNT | Source of data used to compute NNT | Comments |
---|---|---|---|---|---|---|---|---|

Systematic review and meta-analysis | ||||||||

Jonas 2014 | Binary | No | No | Yes | Yes | NNT = 1/RD | Pooled RD | A pooled RD was calculated for two outcomes. Duration of included trials ranged from 12 to 52 weeks for the outcome any drinking, and from 12 to 24 weeks for heaving drinking |

Hempel 2012 | Binary | No | No | Yes | Yes | NNT = 1/RD | Pooled RD | The pooled RD (obtained from meta-analysis) led to a loss of follow-up time. Most trials either did not specify the follow-up period, or the assessment was explicitly limited to the time of antibiotics treatment |

Leucht 2012 | Binary | Yes | Yes | Yes | Yes | NNT = 1/RD | Pooled RD | The outcome is assessed between 7 and 12 months of follow-up; a mean study duration is indicated for each outcome with NNT calculated from absolute RD pooled from the meta-analysis |

Shah 2012 | Binary | No | No | Yes | Yes | NNT = 1/RD | Pooled RD | The study comprehends the calculation and comparison of NNT for several treatments. However, NNTs are not comparable because they were calculated from pooled RDs and times of follow-up vary considerably across studies included in the meta-analysis (10 days to 48 weeks) |

Preiss 2011 | Binary | Yes | Yes | No | No | NNT = 1/RD | Pooled RD | The variable for the primary outcome of the study is binary, and pooled OR (95% CI) was calculated. However, NNT was calculated by taking the reciprocal of RD between pooled event rates per 1000 patient-years. Person-time-based NNT was presented and interpreted as the number of persons needed to treat over 1 year |

Shamliyan 2011 | Binary | Yes | No | Yes | Yes | NNT = 1/RD | Pooled RD | Several antiviral treatments were compared based on estimates of NNT. However, studies with different times of follow-up for antiviral treatments were used to pool absolute RD. The time horizon factor is lost |

Coker 2010 | Binary | Yes | Yes | Yes | No | NNT = 1/RD | Pooled RD | The pooled RD was obtained for a 14 day follow-up duration in all studies included in the meta-analysis. However, RD varies considerably across the studies included in the meta-analysis (ranging from −8% to 27%) |

Testa 2008 | Binary | No | No | Yes | Yes | NNT = 1/RD | Pooled RD | Pooled RD was used to calculate NNT. The follow-up of included studies ranged from ”in hospital” to 6 months |

Bridge 2007 | Binary | Yes | No | Yes | Yes | NNT = 1/RD | Pooled RD | DerSimonian and Laird random-effects model was used to obtain a pooled estimate of the RD (95% CI). NNT was calculated as the reciprocal of RD. The duration of follow-up and the baseline risk varied considerably across included studies |

Dentali 2007 | Binary | Yes | No | No | Yes | NNT = 1/RD | Simple proportions | Raw totals of patients from each study were added together to estimate proportions and calculate RD, i.e., treating data as if all were from one study (Simpson’s paradox). Further, the baseline risk ranged considerably across included studies (e.g., 0.2–4.0% for pulmonary embolism) |

Rovers 2006 | Binary | Yes | Yes | No | No | NNT = 1/RD | Pooled RD | Although it is not clearly stated in the methods section, the discussion of the study suggests that the authors calculated pooled RD by means of the meta-analysis |

Bongartz 2006 | Binary | No | Yes | Yes | Yes | NNT = 1/RD | Pooled RD | NNT calculated for treatment periods of 6–12 months and 3–12 months, using Mantel-Haenszel fixed estimate of absolute RD in cases in which an OR of at least 1.5 was detected |

Spiegel 2006 | Binary | No | No | No | Yes | NNT = 1/RD | Pooled RD | A pooled RD was calculated for two comparisons. Duration of included trials ranged from 6 to 78 weeks for one comparison and from 12 to 24 weeks for another comparison |

Randomized controlled trial | ||||||||

Shepherd 2008 | Time to event | Yes | Yes | No | No | NNT = 1/RD | Simple proportions | NNT calculated as 1/RD using final rates of event and citing a median time of follow-up of 4.8 years (NNT = 14 in patients with diabetes and chronic kidney disease). However, a Kaplan-Meier curve is provided in the study, which should have been used (since the median follow-up is lower than the 5-years objective, at least some patients did not complete the follow-up). From the Kaplan-Meier curve, we would have 20.3% and 14.0% patients with the outcome in the atorvastatin 10 mg and 80 mg/day, respectively, at 4.8 years of follow-up and an NNT = 15.8 |

Retrospective cohort study | ||||||||

Graham 2010 | Time to event | Yes | Yes | Yes | Yes | NNT = 1/RD | Simple proportions | NNT was calculated using RD between unadjusted incidence rates. Adjusted incidence rates from the Kaplan-Meier curves should have been used. For example, at 1 year of follow-up, NNT for the composite endpoint would be 92 from Kaplan-Meier curves, rather than 60 person-years from unadjusted incidence rates. The authors interpreted person-years as number of persons treated over 1 year, which is not exactly the same |