Creatine Cognition Meta-analysis

Does creatine increase cognitive performance? Maybe for vegetarians but probably not.
psychology, meta-analysis, R, IQ, bibliography
2013-09-062019-01-29 finished certainty: unlikely importance: 5


I at­tempt to meta-an­a­lyze con­flict­ing stud­ies about the cog­ni­tive ben­e­fits of cre­a­tine sup­ple­men­ta­tion. The wide va­ri­ety of psy­cho­log­i­cal mea­sures by uni­formly small stud­ies ham­pers any ag­gre­ga­tion. 3 stud­ies mea­sured IQ and turn in a pos­i­tive re­sult, but sug­ges­tive of veg­e­tar­i­an­ism caus­ing half the ben­e­fit. Dis­cus­sions in­di­cate that pub­li­ca­tion bias is at work. Given the va­ri­ety of mea­sures, small sam­ple sizes, pub­li­ca­tion bi­as, pos­si­ble mod­er­a­tors, and smal­l­-s­tudy bi­as­es, any fu­ture cre­a­tine stud­ies should use the most stan­dard mea­sures of cog­ni­tive func­tion like RAPM in a rea­son­ably large pre-reg­is­tered ex­per­i­ment.

is a chem­i­cal found through­out the body in a num­ber of roles; it is most fa­mous for its pres­ence in mus­cles and en­abling greater ex­er­tion, but it also plays a role in the ner­vous sys­tem. Some small psy­chol­ogy ex­per­i­ments in healthy adults have found cog­ni­tive ben­e­fits to sup­ple­men­ta­tion but oth­ers dis­agree (, Ex­am­ine.­com), and these differ­ences may be due to co­vari­ates like be­ing veg­e­tar­ian & hence cre­atine-d­e­fi­cient.

When small stud­ies con­flict, one way to get an­swers is to try to meta-an­a­lyze them into a sin­gle more ro­bust sum­ma­ry. In par­tic­u­lar, I am in­ter­ested in whether cre­a­tine sup­ple­men­ta­tion in­creases IQ. One prob­lem here is that the stud­ies may not use enough of the same mea­sures to in­clude in the same meta-analy­sis; be­sides that, the stud­ies are likely se­verely un­der­pow­ered to de­tect plau­si­ble effects: IQ in­creases have been often claimed, but have rarely panned out (see, for ex­am­ple, ) and “ex­tra­or­di­nary claims re­quire ex­tra­or­di­nary proof”.

Background

While cre­a­tine is fa­mous for its ath­letic uses1, there is also bi­o­log­i­cal ev­i­dence sug­gest­ing cre­a­tine is in­volved in men­tal per­for­mance, serv­ing as a fast source of en­er­gy, cre­atine-re­lated re­tar­da­tion & dis­abil­i­ty, cor­re­lates be­tween meat-eat­ing and per­for­mance etc. For one sur­vey, and a more de­tailed dis­cus­sion of the ra­tio­nale for ex­pect­ing cre­a­tine to help, see Lit­tle­ton 2013; for dis­cus­sion of the neu­ro­pro­tec­tive effects, see Cunha 2017. An in­ter­est­ing lon­gi­tu­di­nal cor­re­la­tion be­tween cre­a­tine lev­els and later in­creased salary is found in Böck­er­man et al 2014. And re­views some an­i­mal ex­per­i­ments sug­gest­ing po­ten­tial ben­e­fits in neu­rode­gen­er­a­tive dis­eases like ag­ing.

Ex­per­i­men­tal­ly, cre­a­tine turns out to boost men­tal per­for­mance in some cir­cum­stances (eg. Ling et al 2009 saw the cre­a­tine group post-s­core 4 points higher than con­trols, or Watan­abe 2002’s less oxy­gen use dur­ing men­tal arith­metic; Jonathan Toomim rec­om­mends it high­ly, claim­ing that “I’m more con­fi­dent that I’ve no­ticed effects [on men­tal per­for­mance] of cre­a­tine than of DnB.”)

These re­sults are a lit­tle mixed. There are stud­ies show­ing ben­e­fits in:

  1. Veg­e­tar­i­ans (Rae 2003)
  2. the sleep­-de­prived (Mc­Mor­ris 2006; Mc­Mor­ris 2007)
  3. the el­derly (Mc­Mor­ris et al 2007)

How­ev­er, Raw­son et al 2008 is a broad null re­sult for healthy om­ni­vores, who are prob­a­bly most of the read­ers of this FAQ. (Jonathan Toomim has crit­i­cized Raw­son et al 2008 as sta­tis­ti­cally weak and us­ing a pos­si­bly not sen­si­tive test of men­tal per­for­mance; oth­ers have pointed out Raw­son et al 2008 ad­min­is­tered a to­tal of cre­atine, which is less than half that of Rae 2003 and 16% less than Ling et al 2009.)

A 2018 meta-analysis/systematic re­view, Avgeri­nos et al 2018, “Effects of cre­a­tine sup­ple­men­ta­tion on cog­ni­tive func­tion of healthy in­di­vid­u­als: A sys­tem­atic re­view of ran­dom­ized con­trolled tri­als”, reached sim­i­lar con­clu­sions as I did in 2013: some ev­i­dence of het­ero­ge­neous ben­e­fits in a lit­er­a­ture which uses too many differ­ent mea­sures.

Overview

The searches & alerts yielded the fol­low­ing po­ten­tially use­ful stud­ies:

We are look­ing for stud­ies of the effects of cre­a­tine on cog­ni­tive per­for­mance in nor­mal healthy pop­u­la­tions un­der nor­mal con­di­tions (ie no dis­eases, no ge­netic dis­or­ders, no ex­otic con­di­tions like hy­pox­i­a).

Us­able stud­ies turn out to em­ploy a wide va­ri­ety of psy­cho­log­i­cal mea­sures:

Study Test
Watan­abe 2002 Uchi­da-Krae­pelin se­r­ial cal­cu­la­tion test
Rae 2003 Raven’s Ad­vanced Pro­gres­sive Ma­tri­ces; Back­ward Digit Span
Mc­Mor­ris 2006 Ran­dom move­ment gen­er­a­tion test; for­ward ver­bal re­call; back­wards ver­bal re­call; 4-choice vi­sual re­ac­tion time test; Hem­mings mood state in­ven­tory
Mc­Mor­ris 2007 Ran­dom num­ber gen­er­a­tion task; num­ber re­call test; four-choice vi­sual re­ac­tion time test; Hem­mings mood state in­ven­to­ry; NASA-TLX effort sub­-s­cale
Mc­Mor­ris et al 2007 Ran­dom num­ber gen­er­a­tion task; for­ward ver­bal re­call; back­wards ver­bal re­call; for­ward Corsi Block Tap­ping test; back­ward Corsi Block Tap­ping test; long-term mem­ory test
Gast­ner et al 2007 Raven’s Ad­vanced Pro­gres­sive Ma­tri­ces; Uchi­da-Krae­pelin se­r­ial cal­cu­la­tion test; Back­ward Digit Span
Raw­son et al 2008 Sim­ple re­ac­tion time; code sub­sti­tu­tion; code sub­sti­tu­tion de­layed; log­i­cal rea­son­ing sym­bol­ic; math­e­mat­i­cal pro­cess­ing; run­ning mem­o­ry; Stern­berg mem­ory re­call
Ling et al 2009 Raven’s Ad­vanced Pro­gres­sive Ma­tri­ces; Mem­ory Scan­ning task; Num­ber-Pair Match­ing task; Sus­tained At­ten­tion task; Ar­row Flanker task
Ham­mett et al 2010 Raven’s Ad­vanced Pro­gres­sive Ma­tri­ces; Back­ward Digit Span
Ben­ton & Dono­hoe 2011 Word re­call; re­ac­tion-time; vig­i­lance rapid in­for­ma­tion pro­cess­ing task; Con­trolled Oral Word As­so­ci­a­tion Test
Alves et al 2013 For­ward Digit Span; Back­ward Digit Span; Mini-Men­tal State Ex­am­i­na­tion; Stroop Test; Trail Mak­ing Test; De­lay Re­call Test
Merege-Filho et al 2016 Stroop Test; Rey Au­di­tory Ver­bal Learn­ing Test; Raven Pro­gres­sive Ma­tri­ces; Trail Mak­ing Test
Cook et al 2011 Rugby ball pass­ing

In to­tal, these stud­ies use ~33 dis­tinct mea­sures of cog­ni­tive func­tion­ing; this het­ero­gene­ity ren­ders any sum­mary diffi­cult as most mea­sures were used in only one ex­per­i­ment, and en­cour­ages se­lec­tive re­port­ing. The Hem­mings mood state in­ven­to­ry, ran­dom num­ber gen­er­a­tion task, for­ward ver­bal re­call, & Uchi­da-Krae­pelin se­r­ial cal­cu­la­tion test are used in 2 ex­per­i­ments each, but the back­ward digit span is used in 4 ex­per­i­ments, and the RAPM is used in 4 ex­per­i­ments - so it would be best to an­a­lyze those two.

RAPM

Data

I de­cided to code mul­ti­ple rel­e­vant vari­ables:

  1. di­et: if no diet was spec­i­fied, as­sume om­niv­o­rous­ness since as lit­tle as 5% of West­ern pop­u­la­tion are veg­e­tar­i­ans

    • 0: om­ni­vore
    • 1: ve­gan or veg­e­tar­ian
  2. sleep:

    • 0: no men­tion is made of sleep de­pri­va­tion
    • 1: if sleep de­pri­va­tion in the ex­per­i­men­tal as op­posed to con­trol group
  3. IQ:

    • 0: RAPM
  4. dose: to­tal amount of cre­a­tine ad­min­is­tered, in grams; av­er­age is to­tal amount of cre­a­tine di­vided by num­ber of days on which cre­a­tine is taken by a sub­ject

  5. age: av­er­age mean of all sub­jects’ age; me­di­ans were treated as means if that was pro­vided in­stead, and means given of range end­points if only that was pro­vided

  6. type: cre­a­tine can be con­sumed in mul­ti­ple forms.

    (CM) is the most com­mon, but also used in a study is (CEE). While CEE was de­vel­oped to al­low smaller doses than CM, Kat­seres et al 2009 sug­gests it breaks down far too fast to be effec­tive and so CEE doses may not be 1:1 equiv­a­lent with CM.

year study n.e mean.e sd.e n.c mean.c sd.c type dose to­tal dose av­er­age age diet sleep
2003 Rae 25 13.7 4.1 25 9.7 3.8 mono­hy­drate 210 5 25.59 1 0
2009 Ling 17 120.3 5.945 17 116.1 9.086 ethyl es­ter 75 5 21 0 0
2010 Ham­mett 11 13.45 2.25 11 11.45 3.8 mono­hy­drate 110 15.7 27.59 0 0
2016 Merege-Filho 35 30.4 4.6 32 31.8 3.8 mono­hy­drate 13.74 1.96 11.5 0 0

Results

The 4 stud­ies do not turn in a sta­tis­ti­cal­ly-sig­nifi­cant pos­i­tive re­sult in the ran­dom-effects meta-analy­sis:

Random-Effects Model (k = 4; tau^2 estimator: REML)

tau^2 (estimated amount of total heterogeneity): 0.2731 (SE = 0.3128)
tau (square root of estimated tau^2 value):      0.5226
I^2 (total heterogeneity / total variability):   72.62%
H^2 (total variability / sampling variability):  3.65

Test for Heterogeneity:
Q(df = 3) = 12.8147, p-val = 0.0051

Model Results:

estimate       se     zval     pval    ci.lb    ci.ub
  0.4259   0.3095   1.3761   0.1688  -0.1807   1.0325
A for­est plot of the 4 stud­ies

When veg­e­tar­i­an­ism is used as a co­vari­ate (this ap­plies only to Rae 2003), it is not sta­tis­ti­cal­ly-sig­nifi­cant ei­ther but does lower the es­ti­mate fur­ther (given that Rae 2003 was also the largest effec­t):

Mixed-Effects Model (k = 4; tau^2 estimator: REML)

tau^2 (estimated amount of residual heterogeneity):     0.2034 (SE = 0.3213)
tau (square root of estimated tau^2 value):             0.4510
I^2 (residual heterogeneity / unaccounted variability): 64.29%
H^2 (unaccounted variability / sampling variability):   2.80
R^2 (amount of heterogeneity accounted for):            25.52%

Test for Residual Heterogeneity:
QE(df = 2) = 5.9274, p-val = 0.0516

Test of Moderators (coefficient(s) 2):
QM(df = 1) = 1.5477, p-val = 0.2135

Model Results:

         estimate      se    zval    pval    ci.lb   ci.ub
intrcpt    0.2095  0.3261  0.6423  0.5207  -0.4297  0.8487
diet       0.7865  0.6322  1.2441  0.2135  -0.4526  2.0257

Un­for­tu­nate­ly, with so few stud­ies I can’t in­ves­ti­gate dose mean­ing­fully

Publication bias

The com­mon pub­li­ca­tion bias checks like the fun­nel plot are use­less with 3 stud­ies. As it hap­pens, there is no need to do any hy­poth­e­sis-test­ing here: Ling (per­sonal com­mu­ni­ca­tion 2013) men­tions that 2 stu­dent the­ses were done in­volv­ing cre­a­tine sup­ple­men­ta­tion & cog­ni­tion with ap­par­ently un­in­ter­est­ing re­sults, but did not have any copies and the uni­ver­sity li­brary had not re­tained any; this is prima fa­cie pub­li­ca­tion bias. Hence, we know that the meta-an­a­lytic re­sults are bi­ased up­wards by pub­li­ca­tion bias.

Backward Digit Span

Data

year study n.e mean.e sd.e n.c mean.c sd.c type dose to­tal dose av­er­age age diet
2010 Ham­mett 11 11 Mono­hy­drate 110 15.7 27.59 0
2003 Rae 25 8.5 1.76 25 7.05 1.19 Mono­hy­drate 210 5 25.59 1
2013 Alves 25 3.75 1.0 22 3.25 0.76 Mono­hy­drate 25 4.17 66.8 0

Results

Source

set.seed(7777) # for reproducible numbers
# TODO: factor out common parts of `png` (& make less square), and `rma` calls
library(XML)
creatine <- readHTMLTable(colClasses = c("integer", "factor", rep("numeric", 6), "factor", rep("numeric", 5)),
                          "https://www.gwern.net/Creatine")[[2]]
# install.packages("metafor") # if not installed
library(metafor)

cat("Basic random-effects meta-analysis of all studies:\n")
res1 <- rma(measure="SMD", m1i = mean.e, m2i = mean.c, sd1i = sd.e, sd2i = sd.c, n1i = n.e, n2i = n.c,
            data = creatine); res1

png(file="~/wiki/images/creatine/forest.png", width = 580, height = 580)
forest(res1, slab = paste(creatine$study, creatine$year, sep = ", "))
invisible(dev.off())

cat("Random-effects with vegetarian covariate:\n")
rma(measure="SMD", m1i = mean.e, m2i = mean.c, sd1i = sd.e, sd2i = sd.c, n1i = n.e, n2i = n.c,
    data=creatine, mods = ~ diet)

system(paste('cd ~/wiki/images/creatine/ &&',
             'for f in *.png; do convert "$f" -crop',
             '`nice convert "$f" -virtual-pixel edge -blur 0x5 -fuzz 10% -trim -format',
             '\'%wx%h%O\' info:` +repage "$f"; done'))
system("optipng -o9 -fix ~/wiki/images/creatine/*.png", ignore.stdout = TRUE)

Study details

  • Watan­abe 2002, Mc­Mor­ris 2006, Mc­Mor­ris et al 2007, Mc­Mor­ris 2007, Raw­son et al 2008, Ben­ton & Dono­hoe 2011: ex­cluded for not us­ing a mea­sure of in­tel­li­gence.
  • Gast­ner et al 2007: ex­cluded for lack of nec­es­sary de­tails.

Rae 2003

In this work, we tested the hy­poth­e­sis that oral cre­a­tine sup­ple­men­ta­tion (5g daily for six weeks) would en­hance in­tel­li­gence test scores and work­ing mem­ory per­for­mance in 45 young adult, veg­e­tar­ian sub­jects in a dou­ble-blind, place­bo-con­trolled, cross-over de­sign. Cre­a­tine sup­ple­men­ta­tion had a sig­nifi­cant pos­i­tive effect (p = 0.0001) on both work­ing mem­ory (back­ward digit span) and in­tel­li­gence (Raven’s Ad­vanced Pro­gres­sive Ma­tri­ces), both tasks that re­quire speed of pro­cess­ing.

Forty-five ve­gan or veg­e­tar­ian sub­jects (12 males (me­dian age of 27.5, range of 19-37 years), 33 fe­males (me­dian age of 24.9, range of 18-40 years); 18 ve­gan (me­dian du­ra­tion of 4.6 years, range of 0.7-17 years) and 27 veg­e­tar­ian (me­dian du­ra­tion of 14.3, range of 1-23 years)) were re­cruited with in­formed con­sent from among the stu­dent pop­u­la­tion of The Uni­ver­sity of Syd­ney

The study fol­lowed a dou­ble-blind, place­bo-con­trolled, cross-over de­sign. Sub­jects were seen on four sep­a­rate oc­ca­sions, at six-week in­ter­vals, fol­low­ing an overnight fast to min­i­mize any fluc­tu­a­tions in blood glu­cose.

A cog­ni­tive test bat­tery was also ad­min­is­tered. At the end of the first and third test ses­sions, sub­jects were given an en­ve­lope marked with their study num­ber and con­tain­ing 5 g doses of sup­ple­ment (cre­a­tine mono­hy­drate ((2-methyl­guanido)acetic acid); Pan Phar­ma­ceu­ti­cals, Aus­tralia) or placebo (mal­todex­trin; Manil­dra Starch­es, Aus­tralia) in plas­tic vials. Sub­jects were asked to con­sume this sup­ple­ment at the same time each day for the next six weeks and re­ceived ad­vice on how best to take this sup­ple­ment to en­sure max­i­mum sol­u­bil­ity and ab­sorp­tion. Sub­jects re­turned the en­ve­lope with un­used vials at the end of each six-week pe­riod and the num­ber of vials re­main­ing was used to as­sess com­pli­ance, val­i­dated against in­creases in red cell (tis­sue) cre­a­tine. Be­tween vis­its 2 and 3, the sub­jects con­sumed no sup­ple­ment. Note: six weeks has been shown to be an ad­e­quate ‘wash-out’ pe­riod (Har­ris et al. 1992).

Sub­jects com­pleted timed (10 min) par­al­lel ver­sions of Raven’s Ad­vanced Pro­gres­sive Ma­tri­ces (RAPMs) con­structed to have equal lev­els of diffi­culty based on the pub­lished nor­ma­tive per­for­mance data and ver­i­fied by us on an in­de­pen­dent sam­ple of 20 sub­jects.

Sup­ple­men­ta­tion with oral cre­a­tine mono­hy­drate sig­nifi­cantly in­creased in­tel­li­gence (as mea­sured by RAPMs done un­der time pres­sure, fig­ure 1a) com­pared with placebo (F3 ,33 = 32.3, p , 0.0001; re­peat­ed-mea­sures ANOVA). There was no sig­nifi­cant effect of treat­ment or­der (F1 ,33 = 1.62, p = 0.21), al­though there was a sig­nifi­cant in­ter­ac­tion with treat­ment or­der (F3 ,99 = 6.7, p = 0.0004). The mean RAPMs raw score un­der placebo was 9.7 (s.d. = 3.8) items cor­rect in 10 min ver­sus 13.7 (s.d. = 4.1) items cor­rect un­der the ex­per­i­men­tal treat­ment. Sup­ple­men­ta­tion with oral cre­a­tine mono­hy­drate (fig­ure 1b) sig­nifi­cantly affected per­for­mance on BDS (F3 ,34 = 29.0, p , 0.0001), with no effect of or­der (F3 ,10 2 = 0.98, p = 0.40). Mean BDS un­der the placebo was 7.05 items (s.d. = 1.19), com­pared with a mean of 8.5 items un­der cre­a­tine treat­ment (s.d. = 1.76).

Gastner et al 2007

“Use of cre­a­tine con­tain­ing prepa­ra­tion e.g. for im­prov­ing mem­o­ry, re­ten­tiv­i­ty, long-term mem­ory and for pre­vent­ing men­tal fa­tigue con­di­tion, com­pris­ing e.g. Ginkgo biloba, gin­seng and niacin” (Eng­lish trans­la­tion, orig­i­nal): Ger­man patent filed in June 2007 by Dr. Thomas Gast­ner, Frauke Selz­er, Dr. Han­s-Peter, Dr. Bendikt Ham­mer, for Alzchem Trost­berg Gmbh:

Use of a cre­a­tine con­tain­ing prepa­ra­tion for im­prov­ing mem­o­ry, re­ten­tiv­i­ty, long-term mem­ory and for pre­vent­ing men­tal fa­tigue con­di­tions, com­pris­ing e.g. at least a fur­ther phys­i­o­log­i­cally effec­tive com­po­nent of the se­ries Ginkgo biloba, gin­seng, taiga root, yam root, lecithin, choline, phos­phatidylser­ine, di­methy­lamino ethanol, acetyl choline, acetyl-L-car­nitine, glu­tathione, glu­t­a­mine, cys­teine, vi­t­a­min A, E, B1, B2, B6, B12, folic acid, pan­tothenic acid and/or zinc, is claimed. Use of a cre­atine-com­po­nent con­tain­ing prepa­ra­tion for im­prov­ing mem­o­ry, re­ten­tiv­i­ty, long-term mem­ory and for pre­vent­ing men­tal fa­tigue con­di­tions, com­pris­ing at least a fur­ther phys­i­o­log­i­cally effec­tive com­po­nent of the se­ries Ginkgo biloba, gin­seng, taiga root, yam root, lecithin, choline, phos­phatidylser­ine, di­methy­lamino ethanol, acetyl choline, acetyl-L-car­nitine, glu­tathione, glu­t­a­mine, cys­teine, vi­t­a­min A, E, B1, B2, B6, B12, E, niac­in, bi­ot­in, folic acid, pan­tothenic acid, zinc, man­gane­se, se­le­ni­um, mag­ne­sium, coen­zyme Q10, glu­cose, colostrum, synephrine, oc­topamine, caffeine, theo­phylline, al­pha -li­nolenic acid, eicos­apen­taenoic acid, omega-3-fatty acid, pirac­etam, anirac­etam, me­man­ti­ne, pyriti­nol, galan­t­a­mine, vin­pocetin, pangamic acid and/or op­tion­ally or­ganic or in­or­ganic salts and/or op­tion­ally es­ters, is claimed.

Sec­tion 2, “Effec­tive­ness”:

Sub­jects were di­vided ran­domly into four groups of 25 peo­ple each. The age of the sub­jects var­ied be­tween 18 and 64 years. The four groups (a-d) were given twice per day for six weeks fol­low­ing each test sub­stances in soft­gel cap­sules:

  1. placebo (1500 mg mal­todex­trin)
  2. cre­a­tine mono­hy­drate (1500 mg)
  3. Ginkgo biloba leaves dry ex­tract (120 mg)
  4. cre­a­tine mono­hy­drate (1500 mg) and Ginkgo biloba leaves dry ex­tract (120 mg)

It mea­sured:

  1. Back­ward Digit Span

    The num­ber of cor­rectly re­peated num­bers be­fore sup­ple­men­ta­tion and the num­ber of cor­rectly re­peated num­bers after six weeks of sup­ple­men­ta­tion are shown in the table. To bet­ter com­pare the re­sults, the differ­ence be­tween the nu­mer­i­cal val­ues is fur­ther il­lus­trat­ed.

    0 weeks 6 weeks Differ­ence
    Placebo (mal­todex­trin) 6.4 6.8 0.4
    Cre­a­tine mono­hy­drate 6.2 7.9 1.7
    Ginkgo biloba 6.7 7.5 0.8
    Cre­a­tine mono­hy­drate + Ginkgo biloba 6.5 9.2 2.7

    Test Method: Wech­sler, D .: Adult In­tel­li­gence Scale man­ual. (1955) New York: Psy­cho­log­i­cal Cor­po­ra­tion.

  2. Raven’s Ad­vanced Pro­gres­sive Ma­tri­ces

    0 weeks 6 weeks Differ­ence
    Placebo (mal­todex­trin) 8.7 10.2 1.5
    Cre­a­tine mono­hy­drate 8.1 12.7 4.6
    Ginkgo biloba 9.8 12.9 3.1
    Cre­a­tine mono­hy­drate + Ginkgo biloba 9.3 17.2 7.9

    Test Method: Rauen, JC et al .: Man­ual for Raven’s pro­gres­sive ma­tri­ces and vo­cab­u­lary scales. (1988) Lon­don: HK Lewis.

  3. Uchi­da-Krae­pelin test

    The test sub­jects a com­pu­ta­tional test was per­formed, which mea­sures the men­tal fa­tigue. They were given sim­ple com­put­ing tasks with an in­ter­val of 5 min­utes twice 15 min­utes. In the sec­ond 15 min­utes, the num­ber of solved prob­lems per minute were de­ter­mined. The test was per­formed be­fore tak­ing sup­ple­men­ta­tion and after 6 weeks. By lin­ear re­gres­sion analy­sis can be in­ferred from the mea­sured data on men­tal fa­tigue. In Ta­ble 3, the re­gres­sion co­effi­cient a is given (). An en­large­ment of the re­gres­sion co­effi­cient is a di­rect mea­sure of a re­duced men­tal fa­tigue.

    0 weeks 6 weeks Differ­ence
    Placebo (mal­todex­trin) -0.0076 -0.0089 -0.0013
    Cre­a­tine mono­hy­drate -0.0088 -0.0046 0.0042
    Ginkgo biloba -0.0105 -0.0081 0.0024
    Cre­a­tine mono­hy­drate + Ginkgo biloba -0.0097 -0.0021 0.0075

    Test Method: Watan­abe, A. et al .: Neu­ro­science Re­search (Ox­ford, United King­dom) (2002), 42 (4), 279-285

But Gast­ner et al re­ported only pre and post-test scores, and not stan­dard de­vi­a­tions; nor was any kind of sta­tis­ti­cal test re­port­ed, mak­ing it diffi­cult to in­fer any­thing about the re­sults.

It is un­clear where this ex­per­i­ment was done, by whom, or whether it was ever pub­lished. (The “in press” ci­ta­tion to Miel­carz et al 2007 turns out to re­fer to Mc­Mor­ris et al 2007.) Noth­ing in the Eng­lish trans­la­tion in­di­cates that it was pub­lished any­where else; searches failed to find any­thing re­lated to this but the patent it­self; the Uchi­da-Krae­pelin test is un­usual and I tried search­ing for any­thing re­lat­ing to it and cre­a­tine in Google/Google-Scholar in Eng­lish & Ger­man but turned up noth­ing be­sides dis­cus­sions of Watan­abe 2002.

On 2013-09-23, I at­tempted to reach Gast­ner via the Alzchem con­tact form. (Gast­ner knows Eng­lish, as demon­strated by co-au­thor­ing “Cre­a­tine - its chem­i­cal syn­the­sis, chem­istry, and le­gal sta­tus” in Cre­a­tine and Cre­a­tine Ki­nase in Health and Dis­ease.) 3 months later on 2013-12-14, I mailed a phys­i­cal let­ter to the Trost­berg, Ger­many ad­dress listed in Cre­a­tine and Cre­a­tine Ki­nase (“De­gussa AG, Dr. Al­bert-Frank-S­traße 32, D-83308 Trost­berg, Ger­many”). As of 2015-01-08, I have re­ceived no re­sponses to any of my at­tempts.

Ling 2009

saw the cre­a­tine group post-s­core 4 points higher than con­trols

There were 34 par­tic­i­pants (in­clud­ing 12 fe­males) who com­pleted the study, with a mean age of 21 years (SD: 1.38; range: 18-24). Par­tic­i­pants were ex­clud­ed, if they pre­sented with a med­ical his­tory of drug and/or al­co­hol abuse, di­ag­nosed psy­chi­atric dis­or­ders, di­a­betes, re­nal in­suffi­ciency (kid­ney dys­func­tion) or had re­cently or were cur­rently sup­ple­ment­ing with a cre­atine-based sub­stance. None of the par­tic­i­pants was veg­e­tar­i­an.

The fi­nal task par­tic­i­pants un­der­took was a mod­i­fied ver­sion of Raven’s Ad­vanced Pro­gres­sive Ma­tri­ces (e.g. Raven et al., 1998) pre­sented on a PC us­ing Macro­me­dia Flash Player. The diffi­culty of the 39 ques­tions grad­u­ally in­creased and was con­strained by a 40-min time lim­it.

The cited iqtest.dk on­line IQ test con­tains only one set of ques­tions and does not ran­dom­ize or vary the se­lec­tion, im­ply­ing that sub­jects an­swered the same ques­tions twice, which is not good (usu­ally IQ tests will come split in equiv­a­lent halves, so one can do pre-tests with the A ques­tions and post-tests with new B ques­tion­s). This may in­val­i­date the ap­par­ent im­prove­ment.

At the end of the first test­ing phase, par­tic­i­pants were given a large en­ve­lope that con­tained 15 plas­tic vials of ei­ther 5 g doses of CEE (ob­tained through the on­line store Dis­count Sup­ple­ments) or a place­bo, mal­todex­trin (ob­tained from the man­u­fac­turer Chem­i­cal Nu­tri­tion; http://www.cnpprofessional.co.uk).

There was a sig­nifi­cant effect of test phase on per­for­mance in the IQ test [F(1,32) = 88.98, P < 0.01] with par­tic­i­pants scor­ing a mean of 112 (SD: 9.44) at base­line, and 118 (7.89) at the end of the study. There was no sig­nifi­cant main effect of sup­ple­ment con­di­tion [F(1,32) = 0.56, NS]. How­ev­er, the in­ter­ac­tion was sig­nifi­cant [F(1,32) = 81.18, P < 0.01]. Pair­wise com­par­isons in­di­cated that par­tic­i­pants in the cre­a­tine con­di­tion per­formed worse than the placebo group in the first phase of test­ing, with base­line means for cre­a­tine group of 108 (SD: 7.42) and for place­bo, 116 (SD: 9.60) (Tukey HSD, P < 0.01). Per­for­mance of the cre­a­tine group also im­proved sig­nifi­cantly over the sup­ple­men­ta­tion pe­ri­od, with the mean of 108 at base­line in­creas­ing to 120 (SD: 5.95) at the end of study (Tukey HSD, P < 0.01). Fur­ther pair­wise com­par­isons in­di­cated that there was no sig­nifi­cant im­prove­ment in the per­for­mance of the placebo group over the sup­ple­men­ta­tion pe­riod (P > 0.05).

Per­for­mance of the cre­a­tine group also im­proved sig­nifi­cantly over the sup­ple­men­ta­tion pe­ri­od, with the mean of 108 at base­line in­creas­ing to 120 (SD: 5.95) at the end of study

Us­ing the spread­sheet of data Ling pro­vided me:

# ling <- read.csv(stdin(),header=TRUE)
Creatine,IQ
1,120
1,118
1,126
1,121
1,119
1,118
1,125
1,117
1,133
1,116
1,124
1,114
1,110
1,130
1,123
1,116
1,115
2,110
2,112
2,105
2,106
2,115
2,105
2,112
2,124
2,119
2,133
2,116
2,123
2,122
2,110
2,125
2,105
2,131
summary(ling)
#     Creatine         IQ
#  Min.   :1.0   Min.   :105
#  1st Qu.:1.0   1st Qu.:112
#  Median :1.5   Median :118
#  Mean   :1.5   Mean   :118
#  3rd Qu.:2.0   3rd Qu.:124
#  Max.   :2.0   Max.   :133
i <- ling[ling$Creatine==1,]$IQ; mean(i); sd(i)
# [1] 120.3
# [1] 5.945
i <- ling[ling$Creatine==2,]$IQ; mean(i); sd(i)
# [1] 116.1
# [1] 9.086
# we use a _t_-test rather than a Wilcoxon to replicate Ling's probable analysis
t.test(IQ ~ Creatine, data=ling)
#     Welch Two Sample t-test
#
# data:  IQ by Creatine
# t = 1.608, df = 27.58, p-value = 0.1192
# alternative hypothesis: true difference in means is not equal to 0
# 95% confidence interval:
#  -1.163  9.634
# sample estimates:
# mean in group 1 mean in group 2
#           120.3           116.1

Hammett 2010

“Di­etary sup­ple­men­ta­tion of cre­a­tine mono­hy­drate re­duces the hu­man fMRI BOLD sig­nal”, Ham­mett et al 2010; quotes rel­e­vant for cal­cu­lat­ing the vari­ables:

To es­tab­lish whether the mag­ni­tude of the BOLD re­sponse is in­flu­enced by Cr lev­els, we have mea­sured re­sponses to vi­sual stim­uli in the pri­mary vi­sual cor­tex (V1) of 22 healthy hu­man vol­un­teers us­ing fMRI, be­fore and after oral ad­min­is­tra­tion of Cr or a placebo (11 in the Cr group and 11 in the placebo group).

The mean and me­dian age of the Cr group was 30.18 and 27 years (SD = 8.37) re­spec­tively and the mean and me­dian age of the placebo group was 25 years (SD = 4.82).

Cre­a­tine sup­ple­men­ta­tion (S­ci-Mx: Glouces­ter­shire, UK) was pro­vided at a dose of 20 g/day for five days, fol­lowed by two ad­di­tional days at a dose of 5 g/day.

In or­der to ver­ify pre­vi­ous re­ports of cog­ni­tive en­hance­ment fol­low­ing Cr sup­ple­men­ta­tion we also mea­sured per­for­mance on the Back­wards Digit Span (BDS) [28] and Raven’s Ad­vanced Pro­gres­sive Ma­tri­ces (RAPM) [24] prior to each scan. The BDS com­prises a set of num­ber se­quences of in­creas­ing length with two differ­ent se­quences of each length. Sub­jects were re­quired to re­peat each se­quence back­wards. The test was ter­mi­nated when the sub­ject failed to re­peat two se­quences of the same length. Differ­ent num­ber se­quences were used for the two test­ing ses­sions. Sub­jects were re­quired to com­plete as many items of the RAPM as pos­si­ble in 5 min. Since the RAPM tests are or­dered in terms of diffi­cul­ty, odd­-num­bered and even-num­bered tests were ad­min­is­tered on weeks 1 and 2 re­spec­tive­ly.

Per­for­mance on the RAPM in­creased non-sig­nifi­cantly by 9.6% fol­low­ing Cr (t = 1.882, df = 10, p = 0.0745) and re­duced non-sig­nifi­cantly by 4.5% (t = 0.7733, df = 10, p = 0.4572) fol­low­ing place­bo. A Group × Week ANOVA re­vealed a main effect of week (F(1, 20) = 5.75, p = 0.026, two-tailed) and a sig­nifi­cant in­ter­ac­tion be­tween week and com­pound (F(1, 20) = 8.58, p = 0.008, two-tailed) for BDS per­for­mance. No sig­nifi­cant effects were found for RAPM per­for­mance.

What’s the stan­dard de­vi­a­tion which pro­duces a p-value of 0.0745 on an in­crease of 9.6% & a sam­ple size of 11 in each group? Hard to tell, but Ham­mett pro­vided me the pre/post scores:

pre post
cre­a­tine mean 12.27 13.45
SD 3.31 2.25
placebo mean 12 11.45
SD 3.52 3.8

Alves et al 2013

Alves data ta­ble

The CR and CR+ST groups re­ceived 20 g of cre­a­tine mono­hy­drate (4 × 5 g/d) for five days fol­lowed by 5 g/d as a sin­gle dose through­out the tri­al.

Alves et al 2013: com­bined the placebo & place­bo+strength­-train­ing groups, and the cre­a­tine & cre­atine+strength­-train­ing groups


  1. The au­thor agrees with cre­atine’s util­ity for ath­let­ics, hav­ing used it for that pur­pose him­self.↩︎