Grandpa - thrice, 47 years as a Statistician in Pharma , over 40 yea promoting Bayesian approaches, 4 years as hobby academic King's Coll London, Cerevisaphile.
I think one point is that old-timers like me used to get closer to the collection of the data than modern day statisticians do. Here are some examples. 1/91. The 1st week I joined the industry I went with my boss and sat in at a clinic of a professor of rheumatology to understand how he ran consultations with patients to make data collection as convenient for him as possible. 2/9
Jun 22, 2020 • 9 tweets • 2 min read
Years ago I headed the non-clinical stats group of a pharma company. We were developing a 3 month version of an existing 1 month therapy & were couldn't manufacture product having the same dissolution profile as the product used for clinical trials. /1
I took a colleague who had worked in aerospace designing industrial experiments to meet scientists from pharmaceutical development. The head of PD told us that there were only 3 factors that influenced dissolution. /2
Mar 1, 2020 • 6 tweets • 2 min read
Years ago I designed a pilot study to test a new experimental pain model. If I remember correctly we were developing anti-rheumatic therapies.
The idea was simple. A subject put their hand into a box. After a random interval a heat source came on. The outcome was the time to withdrawal of the hand. After treatment the same procedure was followed & change in time tolerance was a measure of the effectiveness of therapy.
In early 2001 the FDA approved a treatment for invasive aspergillosis (IA) on the basis of uncontrolled study in 69 patients. At the advisory Committe (AC) the sponsor said:
"Because a randomized controlled study in this disease would be very difficult, data are obtained from a noncomparative study with additional contacts placed by data from the historical control."
In answer to a member of the AC the sponsor went on to say:
Nov 18, 2019 • 4 tweets • 1 min read
The terms "P-value" or "P value" appeared first in the biological literature in the 1910/20s. Harris in The American Naturalist, 1912 wrote of the "value of P" whilst Macarthur in Genetics in 1926, referred to a "P value" directly. Both of these were in a genetics context 1/4
Its appearance in the statistical literature came somewhat later. W Edwards Deming in the "Statistical Adjustment of Data" (1936) used it to describe the results of tests in ANOVAs (log of the variance ratio not the F-test). There are two interesting passages in this extract 2/4
The PRAISE study compared Amlodipine to placebo in CHF. The Primary endpoint was all cause mortality + hospitalisation for major cardiovascular events)
1153 pts (A:582, P:571) randomised – stratified by type of disease ischaemic/non-ischaemic. A 9% reduction in primary endpoint (CI : -24% : +10% - p=0.31).
After the primary analysis, interaction between response and cause of heart failure was investigated.
