Proof of Fabricated Files in a Vitamin C trial by Paul E Marik et al.

 Below is an email I gather sent to Sentara Norfolk Frequent Sanatorium, the editor of CHEST Journal, and prof Paul E Marik. As repeatedly an allegation against a multi-creator paper is no longer an allegation against any particular person creator. 

cc: The Board of Norfolk Sentara Norfolk Frequent Sanatorium

cc: The Compliance officer of Sentara Norfolk Frequent Sanatorium

cc: Prof Paul E Marik

Pricey Editor

Tonight on twitter a paper allegedly describing a 2017 learn about by a “Paul Marik” and physique of workers at Sentara Norfolk Frequent Sanatorium describing a spacious survival rob pleasure in Vitamin C used to be delivered to my attention as a clinical research finding that used to be no longer replicated in extra studies and later reversed. The person said that to their recordsdata no proof of fraud within the behavior of this learn about had been identified. The learn about is chanced on at https://doi.org/10.1016/j.chest.2016.11.036

Sadly within about 5 minutes of reading the learn about it grew to develop into overwhelmingly clear that it’s indeed research fraud and the guidelines is fabricated.

While veritably I’d employ cautious language of “absorbing” or “surprising” patterns within the guidelines and portray “irregularities” and “misfortune”; no such caution is warranted on this case. Right here’s frankly fearless fraud. I gather no longer requested entry to the raw recordsdata or contacted the authors for clarification as the case is fearless no other clarification is most likely.

Allow me to display conceal.

This learn about allegedly describes a sooner than and after learn about, examining the originate of a recent medication regime basically based entirely on vitamin c on mortality in sepsis, claiming a roughly ten fold low cost in loss of life. Every cohort had exactly 47 patients and the patients had been no longer matched. We know this no longer genuine because matching used to be no longer talked about nonetheless because the authors specify that these had been two cohorts of “consecutive” patients, precluding patient matching by definition.

Per this we might maybe perchance quiz that if there used to be no systemic bias the p values for variations in dichotomous baseline characteristics (gender, demographics, comorbidities, diagnoses and many others) would, pointless to claim, centre on 0.5. If systemic variations existed nonetheless the groups might maybe perchance maybe very properly be less equal and p values might maybe perchance maybe merely are seemingly to numbers below 0.5, and this wouldn’t be suspicious in a non-randomised learn about. Systemic biases to p values better than 0.5 aren’t most likely without matching (or some very rare pseudo-block designs no longer relevant right here) except within the atmosphere of fraud.

Sadly every single predominant evaluation (Pneumonia, Urosepsis, Foremost Bacteremia, GI/Biliary, “Different”) is matched perfectly with a p cost from a Fisher unswerving test of 1. 

Indeed of the 23 discrete/continuous variables reported in desk 1 the majority gather a p cost of 1.

The following variables had a p cost of 1, i.e. had been disbursed perfectly evenly for the duration of the two time sessions (or as maximally perfectly evenly as most likely within the case of absorbing numbers).

Acute Kidney Smash: 61 patients

Vasopressors: 44 patients

Foremost Diagnosis – Pneumonia: 37 patients

Coronary heart Failure: 31 patients

Certain blood cultures: 26 patients

Foremost Diagnosis – Urosepsis: 21 patients

COPD: 15 patients

Continual Renal Failure: 15 patients

Foremost Diagnosis – Foremost Bacteremia: 14 patients

Foremost Diagnosis – GI/Biliary: 12 patients

Drug addiction: 10 patients

Foremost Diagnosis – Different: 10 patients

Lack of comorbidities: 3 patients

A more fulsome desk of p values is annexed to the close of this criticism, nonetheless this settle reveals the plotted p values for all 23 dichotomous variables in desk 1. Present that in region of the even distribution between 0 and 1 (or clustering below 1 if the cohorts had been dissimilar) as a exchange most variables gather a p cost of 1, 21/23 variables gather p values over 0.5, and no p cost of below 0.4 is viewed for any of the 23 variables.

This in reality items a little topic in estimating how unlikely these outcomes are, as basically the most traditional test for fraud on this misfortune incessantly is the Stouff-Fisher, nonetheless this is in a position to maybe perchance maybe merely expose these outcomes infinitely unlikely (as the majority of variables gather a p cost of exactly 1), when genuinely it’s maybe more seemingly that it’s trillions to quadrillions to at least one.

A “snappily and dirty” draw to assess this might be to take into story that the possibility of any one variable having a p cost over 0.4 is 60%, the binomial probability of 23 such measures having no seen values beneath 0.4 is (0.6)^23 or no longer as much as 1 in 100,000, this assumes independence which is maybe a microscopic bit unfair, nonetheless seemingly massively underestimates the improbability of such a finding as the outcomes aren’t evenly disbursed between 0.4 and 1.

______