Skip to main content

Table 2 Participants obtaining 30% and 50% improvement in outcomes: risk difference (RD) and number needed to treat (NNT)*

From: Illustrating risk difference and number needed to treat from a randomized controlled trial of spinal manipulation for cervicogenic headache

Improvement criterion SMT
(n = 36)
LM
(n = 39)
adjusted RD
(95% CI)
P adjusted NNT
(95% CI worst to best)
Cervicogenic headache pain scale †
30% 12 wk 47% 38% 20% (2, 37) .025 5.1 (40, 2.7)
  24 wk 58% 49% 17% (2, 33) .028 5.8 (54, 3.1)
50% 12 wk 42% 23% 26% (8, 45) .005 3.8 (13, 2.2)
  24 wk 42% 23% 27% (8, 46) .006 3.8 (13, 2.2)
Cervicogenic headache number (in last 4 wk)
30% 12 wk 72% 51% 23% (6, 40) .007 4.3 (16, 2.5)
  24 wk 78% 62% 14% (1, 27) .040 7.2 (152, 3.7)
50% 12 wk 64% 46% 21% (0, 43) .048 4.7 (442, 2.4)
  24 wk 61% 51% 14% (-2, 30) .094 7.2 (-42, ± ∞, 3.3)
Cervicogenic headache disability scale †
30% 12 wk 64% 51% 15% (-9, 39) ‡ .226 6.7 (-3.5, ± ∞, 2.6)
  24 wk 61% 64% -7% (-29, 15) .553 -15 (-11, ± ∞, 6.5)
50% 12 wk 64% 36% 29% (6, 52) ‡ .015 3.5 (17, 1.9)
  24 wk 56% 38% 20% (-1, 42) .061 4.9 (-105, ± ∞, 2.4)
  1. SMT - spinal manipulative therapy; LM - light massage; NNT - number needed to treat
  2. * Outcomes are presented for the 12-week (short-term) and 24-week (intermediate-term) follow-ups. The SMT and LM group percentages are unadjusted. Missing data were imputed except for five participants with no follow-up data. Differences between groups (risk differences) were adjusted for baseline and all randomization variables. Adjusted NNT = one divided by the adjusted difference between groups. Positive numbers favor spinal manipulation. For the NNT CIs, the limit most favorable to manipulation is on the right and least favorable on the left. Note that for statistically insignificant results, the RD confidence interval includes zero, so that the NNT confidence interval must include 1/0 = ± ∞. These infinity values are more favorable to SMT than a small negative number and less favorable than a small positive number.
  3. † Modified Von Korff scale (scored from 0 to 100 points before dichotomization).
  4. ‡ Linear least-squares regression used in place of binomial regression.