I’ve heard anecdotes about the effectiveness of hookworm infection for people with severe auto-immune issues. The only RCT data I’ve seen is from some brave Canadians on patients with asthma. I’ll mention the study limitations before mentioning the results so that we can decide in advance how big an update to make. 1) the patients received 10 hookworm larvae under a bandage on their arm, not a fecal transplant, but this had been sufficient to establish infection in previous volunteers 2) there were 16 participants in each arm which “powered” the study to detect a 2x change. Results:
Mean PD20AMP improved in the hookworm group by 1.49 (SD 2.00) DD and in the placebo group by 0.98 (4.02) DD, but the mean difference between these changes was not statistically significant [0.51 DD; 95% confidence interval (CI) −1.79 to 2.80; P=0.65, Table 3].
EDIT: Post facto we see that their power calculation was to detect a shift of 1 in PD20AMP with 80% power with the assumption that the standard deviation of this measurement is 1, but actually the standard deviation observed in the pool is about 3.2 (32 patients, 30 degrees of freedom). This suggests the study was underpowered to detect a shift of 1. E.g. the sample size needed to have 80% power to detect a shift of 1 if the population s.d. is 3.2 is 161 patients per arm. (calculated using: “power.t.test(delta=1,sd=3.2,power=0.8)” in R).
[From a Bayesian point of view, the results of the study data themselves are weak supporting evidence (the observed shift was favorable after all). But now I know I’m at risk of cheating mentally, because I was going to update based on the study findings.]
I’ve heard anecdotes about the effectiveness of hookworm infection for people with severe auto-immune issues. The only RCT data I’ve seen is from some brave Canadians on patients with asthma. I’ll mention the study limitations before mentioning the results so that we can decide in advance how big an update to make. 1) the patients received 10 hookworm larvae under a bandage on their arm, not a fecal transplant, but this had been sufficient to establish infection in previous volunteers 2) there were 16 participants in each arm which “powered” the study to detect a 2x change. Results:
EDIT: Post facto we see that their power calculation was to detect a shift of 1 in PD20AMP with 80% power with the assumption that the standard deviation of this measurement is 1, but actually the standard deviation observed in the pool is about 3.2 (32 patients, 30 degrees of freedom). This suggests the study was underpowered to detect a shift of 1. E.g. the sample size needed to have 80% power to detect a shift of 1 if the population s.d. is 3.2 is 161 patients per arm. (calculated using: “power.t.test(delta=1,sd=3.2,power=0.8)” in R).
[From a Bayesian point of view, the results of the study data themselves are weak supporting evidence (the observed shift was favorable after all). But now I know I’m at risk of cheating mentally, because I was going to update based on the study findings.]