Intentional weight loss = CR?

[Guest guest]

Hi All,

Intentional weight loss equates with CR? No, seems to be the answer, below.

See the pdf-available below Medline citation, Medline URL and full-text

excerpts.

Coffey CS, Gadbury GL, Fontaine KR, Wang C, Weindruch R, DB.

The effects of intentional weight loss as a latent variable problem.

Stat Med. 2005 Mar 30;24(6):941-54.

PMID: 15717333

http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve & db=pubmed & dopt=Abstra\

ct & list_uids=15717333 & query_hl=28

.... although caloric restriction (CR)results in dramatically lower body weight

and

prolongs life in multiple species [6],research is equivocal on the e ect of body

weight per

se on rodent longevity [7,8].Moreover,as in humans,under some

circumstances,weight

loss

appears to be associated with increased mortality rate [9].

Despite the common observation that weight loss in humans is often associated

with

in-

creased mortality rate,questions have been raised about the validity or meaning

of

this nding

[10,11].In particular,since weight loss is often a sign of illness,it has been

pointed out that

investigators need to separate intentional weight loss (IWL)from unintentional

weight loss

(UWL [12–14]).Among people who express no intention to lose weight,it has

generally

been assumed that all weight loss subsequently observed is

unintentional.Complementar-

ily,among people who state that they do intend to lose weight,it has generally

been

as-

sumed that all weight loss subsequently observed is intentional,that is,due

solely

to their

intention.

Investigators have studied the association between weight change in these two

groups

sepa-

rately (e.g.References [15–22]).Among people who are not intending to lose

weight,weight

loss has been consistently associated with increased mortality rate.Among people

who

are

intending to lose weight,some studies show apparent bene ts of weight loss

(e.g.Reference

[17]),some are neutral (e.g.Reference [15]),and some show deleterious e ects

(e.g.Refer-

ence [20]).The overall conclusion that some reviewers have drawn from this

literature is that

IWL is at best not bene cial and may even be harmful with respect to mortality

rate

(e.g.

Reference [23]).

We believe that this conclusion is drawn by inappropriately con ating weight

loss

(or more

generally weight change)among people intending to lose weight with IWL (or

change).More

speci cally,we feel that it is unlikely that any study can truly measure IWL per

se

..Rather,

the data that are collected represent the total weight loss among those

intending to

lose weight

and may represent both IWL and UWL.Herein,under certain assumptions,we:(1)show

that

the association between mortality rate and weight loss among people intending to

lose weight

and between mortality rate and IWL are two di erent things;(2)show that the

association

between IWL and mortality rate is an inherently unobservable entity;(3)derive a

method for

estimating the plausible range of true e ect of IWL on mortality rate if one is

willing to make

a number of restrictive,but perhaps reasonable assumptions;and (4)illustrate the

method by

application to a data set involving middle-age onset CR in mice.The methods

described in

this paper are developed precisely for this situation and allow researchers to

obtain a better

understanding of the true relationship between mortality and IWL (which cannot

be

measured)

based on the relationship between mortality rate and observed weight loss among

those who

do and do not intend to lose weight (which can be measured).

.... 5. AN EXAMPLE

We illustrate our approach with an example drawn from the eld of rodent CR

studies

where

animals are typically followed until all are dead and high quality control can

be

achieved

[25].This study involved two strains (B10C3F1 and C57Bl/6J)of male mice studied

to

see

the in uence of CR started in early middle age on longevity and disease

patterns.We

herein

consider the data obtained for the B10C3F1 strain.In brief,135 mice were fed ad

libitum

until 12 months of age at which point they were randomized,individually

housed,and

pro-

vided an intake of either an amount su cient to maintain body weight (control

—unintentional

condition;C =160 kcal =mouse =wk)or an intake of 90 kcal (restricted

—intentional

condition;

R ).To avoid malnutrition,the mice consumed a diet enriched in content of

protein,vitamins

and minerals so that the intakes of these dietary essentials were matched

between

groups.All

animals were observed until death (i.e.there was no right censoring).

From the time of the last pre-randomization weight measurement (12 months of

age)to

23 months of age,the R animals steadily lost weight,with a mean weight loss of

14

:1g

(SD=5 :20).In contrast,C animals gained an average of 0 :45 g (SD =4 :79).We

regressed

lifespan measured in months on weight change scaled in 5 g units (approximately

1

SD)

and obtained estimates of 1 and 1 equal to 0.546 and & #8722;2

:144,respectively.Thus,as in

humans,UWL was associated with reduced lifespan (i.e.each 5 g of UWL was

estimated

to decrease lifespan roughly 2.1 months)and IWL was associated with a small

increase

in

lifespan (roughly 0.5 months/5 g).We tested models in which baseline weight

(weight

at

month 12),the interaction of baseline weight with weight loss,and weight loss

squared were

included,but none were close to signi cant and were therefore dropped from the

models.The

sample variance of weight loss scaled in 5 g units for the two groups were

ˆ2V |1 =(5 :20 =5)2 =1 :08 and ˆ2W =(4 :79 =5)2 =0 :92 (15)

Using the four estimates obtained from these models and the estimation approach

described

above to examine pairs of values for W;Z and 2

Z ,we produced Figure 1.

As can be seen in Figure 1,the plausible e ect of a 5 g IWL in these mice is

& #8722;0 :5–8

months.However,unless one posits that W;Z ¡ & #8722;0 :60,the estimate of 2

remains

larger

than the estimate of 1.Using the partial correlation to check for non-Grammian

covariance

matrices eliminated many possible values.Speci cally,this suggests that there

almost

has to

be a negative correlation between the amounts of IWL versus UWL.Only four

situations

with

a positive correlation between IWL and UWL satis ed this check.However,these

four

values

appear to be outliers since they suggest an implausible increase of 30 months in

lifespan (a

near doubling of lifespan)associated with IWL.To summarize,Figure 1 demonstrates

that

despite the modest increase in lifespan seen in CR animals in the intentional

condition (i.e.

approximately 0.5 months),the true e ect of a 5 g IWL could be substantially

larger

and

biologically quite meaningful.Moreover,although the range of estimates of 2 is

very

large,

as the value of W;Z chosen decreases,the estimate of 2 decreases,but the

estimated

variance

of Z increases in a complementary fashion such that the overall impact of IWL

may

remain

large when expressed in terms of a per cent variance metric.

.... Table I. Selected human studies of mortality and body weight.

...................................

Study name Reference---No. born </= 1913---Repeated measures of weight---Was

intentionality measured?---Comments

...................................

Framingham Heart Study [26]>3000 Yes No Adults aged 28–62 at entry in

1948–1952.

Biannual examinations. As of 1999, there were 993 surviving participants

Honolulu Heart Program Study [27] 8006 men Yes No American men of Japanese

ancestry born in 1900–1919 and living on Oahu in 1965. Examined on 5 occasions

through 1996

Seven Countries Study [28] ~7000 men Yes No Initiated in 1958, a cohort of 12

467

healthy men aged 40–59 from 7 countries ( Finland, Italy, Greece, Japan, The

Netherlands, United States, Yugoslavia) periodically followed The Gothenburg

Study

[29] ~3000 Yes No Four birth cohorts of 70 year olds born in 1901–1922 in

Gothenburg, Sweden followed periodically from 1971–1992

Minnesota Heart Study ( Twin Cities Prospective Study) [30] 217 men Yes No

Men

aged 45–55 at entry in 1948 were reexamined yearly to 1975 and followed up

through

1983

Cancer Prevention Study I [31] >500 000 Self- reported Yes, only at baseline

Approximately 1 million adults enrolled in 1959 and 1960. Eight follow- up

questionnaires through 1972

Bangor Longitudinal Study of Ageing [32] 597 Yes No Adults in rural Wales

aged

65–99 in 1978. Re- examined 6 times and followed through 1999

Amherst College Study [33] ~2500 No No Amherst College students in years

1861–1900. Height and weight at age 20 linked to mortality that covered a

follow- up

period extending to 1949

Terman ’s Lifecycle Study of children with high ability [34] 1428 Yes No

‘Bright’children ( IQ ~135) aged 10–12 residing in California at entry in 1921.

Restudied at 5–10 year intervals for 70 years ( through 1991)

7.DISCUSSION

The method developed herein has broad applicability.First,in

humans,observational

epidemi-

ology studies are frequently used to assess associations between risk factors

and

outcomes.

In the long run,consideration of changes in body composition,including body

fat,may

be

more important than consideration of changes in body weight [3,11,35–39].Perhaps

most

directly relevant is Reference [36]which provided evidence that the loss of body

fat

condi-

tional on change in body weight was associated with reduced mortality

rate,whereas

loss of

body weight conditional on change in body fat was associated with increased

mortality rate.

Following this,we have opined extensively that body composition rather than

merely

body

weight should be examined in future studies and data collection that will allow

such

analyses

are underway in several ongoing studies.However,at the present time,very few (if

any)

longitudinal studies have high-quality body composition measurements at multiple

points in

time on su ciently large numbers of subjects to produce meaningful results with

respect to

mortality rate.For this reason,weight change will likely remain ‘the coin of the

realm’in

this area of inquiry for the near future and,hence,we have chosen to focus on

weight

in this

paper.

However,studies have been conducted that could examine the di erential e ects of

in-

tentional and unintentional body composition changes on other outcomes of

interest

(e.g.

[40,41]).Therefore,it is noteworthy that the method we have developed can be

applied

to

examine the di erential e ects of intentional change in any variable (Z )versus

unintentional

change in that variable (W )on any other variable (Y ),where only the combined

change

(Z+W )is actually observed.In fact,because our method is currently fully

developed

for

situations in which Y is observed without censoring,it is well suited to

situations

as in

Reference [40]or [41]in which body composition changes are examined for their

putative

e ects on changes on other continuous variables that are measurable in the

short-term such

as cardiorespiratory tness.

It might be tempting to think that this does not apply to randomized clinical

trials

(RCTs)

and that,in fact,rather than modelling IWL as a latent variable,we should just

conduct RCTs.

It is,in part,because of the obvious need to clarify the knowledge regarding the

relationship

between IWL and mortality rate underlying these recommendations that the NIH

invested in

the LOOK AHEAD trial,an RCT of the e ects of weight loss on hard endpoints

[42].It

is well

established that the ideal way to de nitively eliminate confounders is to

randomize

subjects

to levels of the independent variable under study.However,in practice,it is not

possible to

randomize people to di erent degrees of weight loss [11].As Yanovski et al

..[43]stated in

a report of a working group that paved the way for the ongoing national LOOK

AHEAD

Trial,‘Subjects in an RCT could not be randomly assigned to lose or not lose

weight;they

could only be randomly assigned to receive or not receive interventions that

might

result in

weight loss.These interventions,however,might well produce changes in health

status

that are

not due to weight loss.Promotion and maintenance of weight loss through

increased

physical

activity,reduced saturated fat intake,and consumption of large amounts of fruit

and

vegetables

are examples of such interventions.It may appear that one could never infer that

weight loss

itself caused the changes in health status.However,if participants in an RCT

were

randomly

assigned to several interventions that produce weight loss through di erent

mechanisms and

these interventions yielded similar improvements in health status,then the

conclusion that

weight loss was responsible for the improvements in health outcomes may be justi

ed.’

The point of this quotation is clear.Although randomization is generally

considered

the

sin qua non of the true experiment and potentially o ers the strongest causal

inferences

of any available study design,the inferential validity of the RCT refers to the

e

ects of

the independent variable to which subjects are assigned [44].Subjects can be

assigned to

treatments that produce,on average ,particular degrees of weight loss but

because:(a)the

treatments themselves may have e ects beyond the weight loss per se

and;(b)within

any one

treatment condition,there will be uncontrolled variability in weight change,some

of

which

may be due to unintentional factors,RCTs cannot be counted upon to yield

unbiased

estimates

of the e ect of IWL.This is not to be dismissive of the enormous value of RCTs

in

this

area.De nitively testing whether treatments that produce weight loss have bene

cial

e ects is

important in and of itself.The point is that such RCTs will not fully address

questions about

the e ects of IWL.Therefore,even in RCTs,the amounts of IWL are still

unobservable

and

may be confounded to some extent.Appropriate statistical methods are needed for

RCT’s in

which the e ects of post-randomization weight change are estimated and the

approach

o ered

herein can be considered a rst generation of such methods.

Finally,the method we propose can also,as we have shown,be applied to animal CR

experiments.Animal experiments are usually randomized and are essentially just

RCTs

in

model organisms.Thus,our comments about the applicability of our method to RCTs

apply

equally to animal studies.

As mentioned by Yang et al .[11],a more critical variable than whether weight

loss

is

intentional or unintentional may be what more proximal factors produced the

weight

loss

following an intention or lack thereof.People try to lose weight through a wide

variety of

methods [45],some of which,such as increasing cigarette smoking [46],may have

profound

deleterious e ects on mortality rate.On the other hand,people may lose weight by

adopting

a more healthy lifestyle without having any intention of losing weight.These

proximal causes

of weight change may have independent e ects on mortality rate and may also

moderate

the

e ects of the weight change they produce.Therefore,in addition to trying to

separate

the

e ects of IWL from UWL,future studies might bene t from modelling the putative

main

e ects of the proximal causes of weight loss on mortality rate and the

interactions

of these

proximal causes with subsequent weight change with respect to in uencing

mortality rate.

Al Pater, PhD; email: old542000@...

__________________________________________________