Introduction|NNP Older|JJR adults|NNS are|VBP frequently|RB counseled|VBN to|TO lose|VB weight|NN ,|, even|RB though|IN there|EX is|VBZ little|JJ evidence|NN that|DT overweight|NN is|VBZ associated|VBN with|IN increased|VBN mortality|NN in|IN those|DT over|JJ age|NN 65|CD .|. 
Six|CD large|RB controlled|VBN population-based|JJ studies|NNS of|IN non-smoking|JJ older|JJR adults|NNS have|VBP investigated|VBN the|DT association|NN between|IN body|NN mass|NN index|NN (|( BMI|NNP )|) and|CC mortality|NN ,|, controlling|VBG for|IN relevant|JJ covariates|NNS [|NN 1|CD 2|CD 3|CD 4|CD 5|CD 6|CD ]|NN .|. 
All|DT studies|NNS found|VBD excess|JJ risk|NN for|IN persons|NNS with|IN very|RB low|JJ BMI|NNP ,|, but|CC that|IN persons|NNS with|IN moderately|RB high|JJ BMI|NNP had|VBD little|JJ or|CC no|DT extra|JJ risk|NN except|IN in|IN certain|JJ small|JJ subsets|NNS .|. 
A|DT review|NN of|IN 13|CD studies|NNS of|IN older|JJR adults|NNS drew|VBD similar|JJ conclusions|NNS [|NN 7|CD ]|NN .|. 

Many|JJ healthy|JJ older|JJR adults|NNS report|VBP gradual|JJ weight|NN gain|NN throughout|IN adult|NN life|NN .|. 
It|PRP may|MD be|VB that|IN a|DT small|JJ amount|NN of|IN gradual|JJ weight|NN gain|NN is|VBZ normative|JJ and|CC associated|VBN with|IN the|DT most|RBS robust|JJ health|NN as|IN we|PRP age|VBP .|. 
It|PRP has|VBZ been|VBN suggested|VBN that|IN weight|NN standards|NNS be|VB adjusted|VBN upwards|NNS for|IN age|NN [|NN 8|CD ]|NN .|. 
Such|JJ recommendations|NNS remain|VBP controversial|JJ ,|, however|RB ,|, because|IN the|DT number|NN of|IN studies|NNS of|IN older|JJR persons|NNS is|VBZ fairly|RB small|JJ ,|, and|CC because|IN few|JJ studies|NNS have|VBP examined|VBN the|DT relation|NN of|IN BMI|NNP to|TO quality|NN of|IN life|NN or|CC years|NNS of|IN healthy|JJ life|NN (|( YHL|NNP )|) in|IN the|DT elderly|JJ [|NN 9|CD ]|NN .|. 

In|IN older|JJR adults|NNS ,|, risk|NN factors|NNS may|MD have|VB a|DT greater|JJR effect|NN on|IN health|NN than|IN on|IN mortality|NN .|. 
If|IN so|RB ,|, then|RB behavior|NN change|NN trials|NNS of|IN weight|NN modification|NN might|MD be|VB more|RBR successful|JJ if|IN they|PRP were|VBD evaluated|VBN on|IN improved|VBN health|NN ,|, rather|RB than|IN on|IN decreased|VBD mortality|NN .|. 
Clinical|NNP trials|NNS powered|VBN to|TO detect|VB differences|NNS in|IN YHL|NNP would|MD often|RB require|VB fewer|JJR subjects|NNS than|IN trials|NNS to|TO detect|VB survival|NN differences|NNS or|CC cardiovascular|JJ events|NNS [|NN 10|CD ]|NN .|. 
In|IN this|DT paper|NN we|PRP study|VBP whether|IN BMI|NNP at|IN baseline|NN is|VBZ associated|VBN with|IN living|VBG longer|RBR ,|, and|CC /|NN or|CC with|IN more|JJR years|NNS of|IN being|VBG healthy|JJ ,|, in|IN a|DT cohort|NN of|IN older|JJR adults|NNS for|IN whom|WP risk|NN factors|NNS ,|, subclinical|JJ disease|NN ,|, and|CC morbidity|NN are|VBP well|RB characterized|VBN .|. 
The|DT goal|NN is|VBZ to|TO determine|VB whether|IN analyses|NNS based|VBN on|IN years|NNS of|IN life|NN (|( YOL|NNP )|) or|CC on|IN YHL|NNP would|MD provide|VB substantively|RB different|JJ results|NNS ,|, and|CC which|WDT measure|NN would|MD yield|VB more|RBR powerful|JJ evaluations|NNS of|IN weight|NN modification|NN interventions|NNS in|IN older|JJR adults|NNS .|. 

Materials|NNS and|CC methods|NNS Study|NN design|NN :|: The|DT Cardiovascular|NNP Health|NNP Study|NNP The|DT Cardiovascular|NNP Health|NNP Study|NNP (|( CHS|NNP )|) is|VBZ a|DT population-based|JJ longitudinal|NN study|NN of|IN 5,888|CD adults|NNS aged|VBN 65|CD and|CC older|JJR at|IN baseline|NN [|NN 11|CD ]|NN .|. 
Subjects|NNP were|VBD recruited|VBN from|IN a|DT random|JJ sample|NN of|IN the|DT Medicare|NNP eligibility|NN lists|NNS in|IN four|CD US|NNP counties|NNS .|. 
Extensive|NNP baseline|NN data|NNS were|VBD collected|VBN for|IN all|DT subjects|NNS using|VBG a|DT baseline|NN home|NN interview|NN ,|, an|DT annual|JJ mail|NN questionnaire|NN ,|, and|CC annual|JJ clinic|NN examinations|NNS .|. 
Additional|JJ information|NN was|VBD collected|VBN in|IN a|DT brief|JJ telephone|NN interview|NN 6|CD months|NNS after|IN each|DT scheduled|JJ visit|NN .|. 
Two|CD cohorts|NNS were|VBD followed|VBN ,|, one|CD with|IN 7|CD years|NNS of|IN follow-up|JJ (|( n|NN =|SYM 5,201|CD )|) and|CC the|DT second|JJ (|( all|DT African|NNP American|NNP ,|, n|NN =|SYM 687|CD )|) with|IN 4|CD years|NNS of|IN follow-up|JJ to|TO date|NN .|. 
Data|NNP collection|NN began|VBD in|IN 1989|CD ,|, and|CC follow-up|JJ is|VBZ virtually|RB complete|JJ for|IN all|DT surviving|VBG subjects|NNS [|NN 12|CD ]|NN .|. 

Body|NN mass|NN index|NN BMI|NNP was|VBD calculated|VBN as|IN measured|VBN weight|NN in|IN kilograms|NNS divided|VBN by|IN the|DT square|NN of|IN measured|VBN height|NN in|IN meters|NNS .|. 
A|DT report|NN from|IN the|DT National|NNP Heart|NNP Lung|NNP and|CC Blood|NNP Institute|NNP classifies|VBZ normal|JJ weight|NN (|( without|IN reference|NN to|TO age|NN )|) as|IN a|DT BMI|NNP of|IN 18.5|CD to|TO 24.9|CD ;|: overweight|NN as|IN 25|CD to|TO 29.9|CD ;|: and|CC obesity|NN as|IN 30.0|CD and|CC higher|JJR [|NN 13|CD ]|NN .|. 
We|PRP consider|VBP separately|RB the|DT group|NN with|IN BMI|NNP between|IN 18.5|CD and|CC 20|CD ,|, which|WDT was|VBD associated|VBN with|IN lower|JJR survival|NN in|IN studies|NNS cited|VBD above|IN .|. 

Years|NNS of|IN life|NN and|CC years|NNS of|IN healthy|JJ life|NN YOL|NNP is|VBZ the|DT number|NN of|IN years|NNS that|IN a|DT person|NN lived|VBD in|IN the|DT 7|CD years|NNS after|IN baseline|NN .|. 
YHL|NNP is|VBZ the|DT number|NN of|IN years|NNS in|IN which|WDT the|DT person|NN was|VBD '|POS healthy|JJ '|POS ,|, and|CC is|VBZ similar|JJ in|IN concept|NN to|TO quality-adjusted|JJ life-years|JJ ,|, healthy|JJ year|NN equivalents|NNS ,|, or|CC active|JJ life|NN expectancy|NN [|NN 14|CD ]|NN .|. 
We|PRP based|VBD YHL|NNP on|IN self-rated|JJ health|NN (|( is|VBZ your|PRP$ health|NN excellent|JJ ,|, very|RB good|JJ ,|, good|JJ ,|, fair|JJ ,|, or|CC poor|JJ ?|. )|) (|( EVGFP|NNP )|) which|WDT was|VBD collected|VBN every|DT 6|CD months|NNS .|. 
EVGFP|NNP is|VBZ a|DT simple|JJ but|CC well-known|JJ measure|NN ,|, which|WDT has|VBZ been|VBN studied|VBN in|IN detail|NN [|NN 15|CD 16|CD ]|NN ,|, and|CC is|VBZ predictive|JJ of|IN health|NN events|NNS in|IN many|JJ studies|NNS [|NN 17|CD ]|NN .|. 
Because|IN we|PRP are|VBP examining|VBG health|NN status|NN over|IN time|NN ,|, we|PRP added|VBD a|DT sixth|JJ health|NN state|NN ,|, dead|JJ .|. 
Data|NNP were|VBD available|JJ about|IN 93|CD %|NN of|IN the|DT time|NN .|. 
We|PRP used|VBD linear|JJ interpolation|NN to|TO estimate|VB missing|VBG data|NNS when|WRB there|EX were|VBD known|VBN values|NNS before|IN and|CC after|IN the|DT missing|VBG value|NN ,|, bringing|VBG the|DT percent|NN complete|JJ to|TO 95|CD %|NN [|NN 18|CD ]|NN .|. 

For|IN this|DT analysis|NN we|PRP defined|VBD YHL|NNP as|IN the|DT number|NN of|IN years|NNS (|( of|IN 7|CD )|) in|IN which|WDT a|DT person|NN reported|VBD excellent|JJ ,|, very|RB good|JJ ,|, or|CC good|JJ health|NN (|( were|VBD '|POS healthy|JJ '|POS )|) .|. 
YHL|NNP ranges|NNS from|IN 0|CD (|( for|IN persons|NNS who|WP were|VBD never|RB in|IN excellent|JJ ,|, very|RB good|JJ ,|, or|CC good|JJ health|NN )|) to|TO 7|CD years|NNS (|( for|IN persons|NNS who|WP were|VBD healthy|JJ throughout|IN )|) .|. 
Since|IN people|NNS reported|VBD their|PRP$ health|NN every|DT 6|CD months|NNS ,|, YHL|NNP has|VBZ a|DT reasonably|RB continuous|JJ distribution|NN .|. 
A|DT drawback|NN of|IN this|DT simple|JJ definition|NN of|IN '|POS healthy|JJ '|POS is|NNS that|IN it|PRP does|VBZ not|RB distinguish|VB between|IN fair|JJ or|CC poor|JJ health|NN and|CC death|NN ,|, since|IN all|DT are|VBP considered|VBN '|POS not|RB healthy|JJ '|'' .|. 
We|PRP also|RB used|VBD an|DT alternative|JJ approach|NN ,|, which|WDT assigns|VBZ a|DT different|JJ value|NN to|TO each|DT level|NN of|IN EVGFP|NNP [|NN 19|CD ]|NN .|. 
Preliminary|JJ results|NNS were|VBD similar|JJ for|IN the|DT two|CD approaches|NNS ,|, however|RB ,|, and|CC we|PRP report|VBP results|NNS using|VBG only|RB the|DT simpler|JJR definition|NN .|. 

The|DT calculations|NNS had|VBD to|TO be|VB modified|VBN to|TO include|VB the|DT 438|CD persons|NNS in|IN the|DT second|JJ African|NNP American|JJ cohort|NN ,|, who|WP have|VBP been|VBN followed|VBN only|RB 4|CD years|NNS to|TO date|NN .|. 
For|IN those|DT persons|NNS ,|, and|CC for|IN 70|CD persons|NNS in|IN the|DT first|JJ cohort|NN who|WP did|VBD not|RB have|VB complete|JJ data|NNS ,|, we|PRP estimated|VBD the|DT last|JJ 4|CD years|NNS of|IN YOL|NNP and|CC YHL|NNP from|IN their|PRP$ age|NN ,|, sex|NN ,|, and|CC health|NN at|IN the|DT end|NN of|IN 3|CD years|NNS ,|, using|VBG validated|JJ methods|NNS presented|VBN elsewhere|RB [|NN 20|CD ]|NN .|. 
That|DT article|NN showed|VBD that|WDT estimated|VBD 4|CD -|: year|NN YOL|NNP and|CC YHL|NNP were|VBD unbiased|JJ for|IN the|DT African|NNP American|JJ cohort|NN .|. 
In|IN the|DT primary|JJ analysis|NN we|PRP used|VBD observed|VBD 7|CD -|: year|NN YOL|NNP and|CC YHL|NNP when|WRB they|PRP were|VBD available|JJ ,|, and|CC observed|VBD 3|CD -|: year|NN YOL|NNP and|CC YHL|NNP plus|CC 4|CD -|: year|NN estimated|VBD YOL|NNP and|CC YHL|NNP when|WRB they|PRP were|VBD not|RB (|( about|IN 10|CD %|NN of|IN the|DT sample|NN )|) .|. 
We|PRP performed|VBD all|DT analyses|NNS with|IN and|CC without|IN the|DT persons|NNS who|WP had|VBD partially|RB estimated|VBN data|NNS ,|, to|TO ensure|VB that|IN the|DT estimation|NN had|VBD not|RB distorted|JJ the|DT findings|NNS .|. 

Covariates|NNP The|DT goal|NN is|VBZ to|TO examine|VB the|DT association|NN of|IN YOL|NNP and|CC YHL|NNP with|IN BMI|NNP .|. 
To|TO adjust|VB for|IN possible|JJ confounding|VBG we|PRP chose|VBD baseline|NN covariates|NNS that|WDT were|VBD prevalent|JJ in|IN the|DT elderly|JJ ,|, related|VBN to|TO mortality|NN and|CC morbidity|NN in|IN previous|JJ studies|NNS ,|, and|CC likely|JJ to|TO be|VB related|VBN to|TO BMI|NNP .|. 
Self-reported|NNP covariates|NNS include|VBP age|NN ,|, gender|NN ,|, smoking|NN (|( never|RB or|CC former|JJ )|) ,|, history|NN of|IN arthritis|NN ,|, cancer|NN ,|, diabetes|NN ,|, fair|JJ or|CC poor|JJ self-rated|JJ health|NN status|NN ,|, limitations|NNS in|IN activities|NNS of|IN daily|JJ living|NN or|CC in|IN instrumental|JJ activities|NNS of|IN daily|JJ living|NN ,|, and|CC 10|CD pounds|NNS or|CC more|RBR unintended|JJ weight|NN loss|NN in|IN the|DT year|NN before|IN baseline|NN .|. 
Clinical|NNP covariates|NNS include|VBP hypertension|NN ,|, cardiovascular|JJ disease|NN (|( prevalent|JJ heart|NN disease|NN ,|, peripheral|JJ vascular|NN disease|NN ,|, or|CC cerebrovascular|NN disease|NN )|) ,|, maximum|NN thickness|NN of|IN the|DT internal|JJ carotid|NN artery|NN ,|, depression|NN (|( CESD|NNP score|NN )|) ,|, serum|NN albumin|NN ,|, serum|NN cholesterol|NN ,|, and|CC serum|NN creatinine|NN .|. 
These|DT measures|NNS are|VBP explained|VBN in|IN more|JJR detail|NN elsewhere|RB [|NN 21|CD 22|CD 23|CD 24|CD ]|NN .|. 
We|PRP excluded|VBD 697|CD current|JJ smokers|NNS and|CC 313|CD others|NNS with|IN incomplete|JJ covariate|NN data|NNS ,|, leaving|VBG 4,878|CD persons|NNS on|IN whom|WP this|DT analysis|NN is|VBZ based|VBN .|. 

Analysis|NNP All|DT analyses|NNS were|VBD performed|VBN separately|RB for|IN men|NNS and|CC women|NNS .|. 
We|PRP calculated|VBD two|CD sets|NNS of|IN adjusted|VBN values|NNS ,|, as|IN follows|VBZ .|. 
We|PRP regressed|JJ YOL|NNP and|CC YHL|NNP first|JJ on|IN age|NN ,|, age|NN squared|VBD ,|, race|NN ,|, and|CC smoking|NN history|NN (|( former|JJ or|CC never|RB )|) ,|, and|CC second|JJ on|IN all|DT of|IN the|DT covariates|NNS listed|VBN above|IN .|. 
We|PRP calculated|VBD adjusted|VBN YOL|NNP as|IN a|DT person|NN 's|POS observed|VBD YOL|NNP minus|CC predicted|VBD YOL|NNP (|( from|IN the|DT regression|NN )|) plus|CC the|DT mean|NN YOL|NNP (|( 6.52|CD years|NNS for|IN women|NNS or|CC 6.06|CD for|IN men|NNS )|) .|. 
That|DT is|VBZ ,|, a|DT person|NN 's|POS adjusted|VBN YOL|NNP is|VBZ his|PRP$ residual|JJ from|IN the|DT regression|NN plus|CC the|DT grand|JJ mean|NN .|. 
The|DT mean|NN of|IN this|DT new|JJ variable|NN ,|, for|IN a|DT group|NN of|IN subjects|NNS ,|, is|VBZ the|DT adjusted|JJ mean|NN YOL|NNP for|IN that|DT group|NN .|. 
Adjusted|NNP YHL|NNP was|VBD calculated|VBN in|IN a|DT similar|JJ manner|NN .|. 
We|PRP calculated|VBD two|CD sets|NNS of|IN adjusted|VBN variables|NNS because|IN of|IN the|DT possibility|NN of|IN '|POS over-adjustment|JJ '|POS ,|, controlling|VBG inappropriately|RB for|IN factors|NNS (|( such|JJ as|IN diabetes|NN )|) which|WDT may|MD have|VB been|VBN causally|RB affected|VBN by|IN the|DT person|NN 's|POS weight|NN .|. 
We|PRP plotted|VBD mean|VB adjusted|VBN YOL|NNP and|CC YHL|NNP against|IN BMI|NNP ,|, and|CC tested|VBD for|IN difference|NN among|IN BMI|NNP groups|NNS using|VBG confidence|NN intervals|NNS or|CC analysis|NN of|IN variance|NN .|. 
Finally|RB we|PRP calculated|VBD the|DT effect|NN size|NN for|IN each|DT measure|NN ,|, comparing|VBG each|DT BMI|NNP subgroup|NN to|TO the|DT '|POS normal|JJ '|POS group|NN .|. 
The|DT effect|NN size|NN is|VBZ the|DT difference|NN in|IN mean|NN YOL|NNP (|( or|CC YHL|NNP )|) in|IN two|CD groups|NNS divided|VBN by|IN their|PRP$ common|JJ standard|JJ deviation|NN .|. 
Since|IN the|DT sample|NN size|NN required|VBN to|TO detect|VB an|DT effect|NN of|IN this|DT magnitude|NN is|VBZ proportional|JJ to|TO the|DT inverse|NN of|IN the|DT squared|JJ effect|NN size|NN ,|, large|JJ effect|NN sizes|NNS are|VBP desirable|JJ .|. 

Results|NNS Table|NNP 1|CD shows|VBZ the|DT distribution|NN of|IN key|JJ variables|NNS by|IN sex|NN and|CC race|NN .|. 
Mean|VB age|NN at|IN baseline|NN was|VBD 73.1|CD and|CC about|IN two|CD thirds|NNS of|IN the|DT men|NNS and|CC a|DT third|JJ of|IN the|DT women|NNS were|VBD former|JJ smokers|NNS .|. 
Black|JJ women|NNS had|VBD a|DT higher|JJR mean|NN BMI|NNP and|CC higher|JJR percent|NN obese|NN (|( BMI|NNP ≥|NN 30|CD )|) than|IN the|DT other|JJ three|CD groups|NNS .|. 
Black|JJ men|NNS were|VBD most|RBS likely|JJ to|TO have|VB unintentionally|RB lost|VBN more|JJR than|IN 10|CD pounds|NNS in|IN the|DT past|JJ year|NN ;|: white|JJ women|NNS were|VBD least|JJS likely|JJ .|. 

About|IN 78|CD %|NN of|IN the|DT subjects|NNS were|VBD healthy|JJ at|IN baseline|NN ,|, declining|VBG to|TO 57|CD %|NN at|IN the|DT end|NN of|IN 7|CD years|NNS ;|: 20|CD %|NN had|VBD died|VBN (|( data|NNS not|RB shown|VBN )|) .|. 
Of|IN the|DT 22|CD %|NN who|WP were|VBD unhealthy|JJ (|( fair|JJ or|CC poor|JJ )|) at|IN baseline|NN ,|, about|IN 24|CD %|NN were|VBD healthy|JJ 7|CD years|NNS later|RB .|. 
There|EX was|VBD thus|RB substantial|JJ change|NN in|IN EVGFP|NNP over|IN time|NN ,|, in|IN both|DT directions|NNS .|. 
Table|NNP 1|CD shows|VBZ the|DT mean|NN YOL|NNP and|CC YHL|NNP (|( calculated|VBN from|IN EVGFP|NNP )|) in|IN the|DT first|JJ seven|CD years|NNS of|IN the|DT study|NN ,|, adjusted|VBN to|TO age|NN 73|CD .|. 
For|IN example|NN ,|, black|JJ women|NNS averaged|VBD 6.3|CD YOL|NNP ,|, but|CC only|RB 4.2|CD YHL|NNP of|IN a|DT maximum|NN possible|JJ 7|CD .|. We|PRP calculated|VBD some|DT additional|JJ descriptive|JJ statistics|NNS ,|, shown|VBN in|IN the|DT final|JJ two|CD lines|NNS :|: years|NNS of|IN unhealthy|JJ life|NN (|( YOL|NNP minus|CC YHL|NNP )|) and|CC years|NNS lost|VBD to|TO death|NN (|( 7|CD minus|CC YOL|NNP )|) .|. 
White|JJ women|NNS had|VBD the|DT most|JJS YHL|NNP and|CC black|JJ men|NNS the|DT fewest|JJS ;|: black|JJ women|NNS had|VBD the|DT most|JJS years|NNS of|IN unhealthy|JJ life|NN ,|, and|CC white|JJ men|NNS the|DT fewest|JJS ;|: black|JJ men|NNS lost|VBD the|DT most|JJS years|NNS to|TO death|NN (|( 1.3|CD out|IN of|IN 7|CD )|) while|IN white|JJ women|NNS lost|VBD only|RB 0.4|CD years|NNS .|. 
For|IN blacks|NNS ,|, about|IN 68|CD %|NN of|IN their|PRP$ YOL|NNP were|VBD healthy|JJ (|( YHL|NNP /|NN YOL|NNP ,|, not|RB shown|VBN )|) ;|: for|IN whites|NNS ,|, about|IN 75|CD %|NN were|VBD healthy|JJ .|. 

Among|IN whites|NNS ,|, the|DT gender|NN differences|NNS in|IN Table|NNP 1|CD were|VBD statistically|RB significant|JJ (|( p|NN <|NN .05|CD )|) except|IN for|IN BMI|NNP and|CC unintended|JJ weight|NN loss|NN .|. 
Among|IN blacks|NNS ,|, gender|NN differences|NNS were|VBD significant|JJ except|IN for|IN 10|CD pounds|NNS unintended|JJ weight|NN loss|NN and|CC weight|NN loss|NN since|IN age|NN 50|CD .|. 
Among|IN males|NNS ,|, there|EX were|VBD significant|JJ differences|NNS between|IN black|NN and|CC white|JJ for|IN BMI|NNP ,|, unintended|JJ weight|NN loss|NN ,|, YOL|NNP ,|, YHL|NNP ,|, years|NNS of|IN unhealthy|JJ life|NN ,|, and|CC years|NNS lost|VBD to|TO death|NN .|. 
Whites|NNP in|IN the|DT sample|NN had|VBD higher|JJR income|NN and|CC education|NN (|( data|NNS not|RB shown|VBN )|) .|. 
After|IN adjusting|VBG for|IN income|NN and|CC education|NN ,|, as|RB well|RB as|IN age|NN and|CC former|JJ smoking|NN ,|, the|DT difference|NN in|IN BMI|NNP was|VBD no|RB longer|RB statistically|RB significant|JJ .|. 
Among|IN females|NNS ,|, blacks|NNS and|CC whites|NNS differed|VBD significantly|RB on|IN BMI|NNP ,|, BMI|NNP >|NN 30|CD ,|, weight|NN loss|NN since|IN age|NN 50|CD ,|, YOL|NNP ,|, YHL|NNP ,|, years|NNS of|IN unhealthy|JJ life|NN ,|, and|CC years|NNS lost|VBD to|TO death|NN .|. 
After|IN adjustment|NN for|IN income|NN and|CC education|NN ,|, the|DT difference|NN in|IN weight|NN loss|NN since|IN age|NN 50|CD was|VBD no|DT longer|RB significant|JJ .|. 
Blacks|NNS had|VBD significantly|RB lower|JJR YOL|NNP and|CC YHL|NNP than|IN whites|NNS after|IN adjustment|NN for|IN age|NN ,|, but|CC the|DT difference|NN disappeared|VBD after|IN adjustment|NN for|IN the|DT entire|JJ set|NN of|IN health-related|JJ baseline|NN covariates|NNS (|( analyses|NNS not|RB shown|VBN )|) .|. 

We|PRP next|RB examined|VBD the|DT relationship|NN of|IN BMI|NNP to|TO YOL|NNP and|CC YHL|NNP .|. 
Table|NNP 2|CD presents|VBZ the|DT mean|NN values|NNS of|IN YOL|NNP and|CC YHL|NNP ,|, adjusted|VBN for|IN age|NN ,|, race|NN ,|, and|CC previous|JJ smoking|NN (|( columns|NNS 1|CD and|CC 3|CD )|) ,|, and|CC also|RB adjusted|VBN for|IN the|DT entire|JJ set|NN of|IN covariates|NNS (|( columns|NNS 2|CD and|CC 4|CD )|) .|. 
For|IN example|NN ,|, YOL|NNP for|IN women|NNS ,|, adjusted|VBN for|IN age|NN ,|, race|NN ,|, and|CC smoking|NN ,|, averaged|VBD 6.0|CD years|NNS for|IN women|NNS with|IN a|DT baseline|JJ BMI|NNP below|IN 18.5|CD ,|, but|CC averaged|VBD 6.6|CD years|NNS for|IN women|NNS with|IN a|DT BMI|NNP from|IN 25|CD to|TO 29.9|CD .|. The|DT second|JJ column|NN ,|, which|WDT shows|VBZ results|NNS adjusted|VBN for|IN all|DT covariates|NNS ,|, is|VBZ not|RB very|RB different|JJ (|( the|DT only|JJ discrepancy|NN is|VBZ for|IN men|NNS with|IN BMI|NNP <|NN 18.5|CD ,|, a|DT category|NN containing|VBG only|RB 14|CD men|NNS )|) .|. 
Adjustment|NNP for|IN extensive|JJ covariates|NNS also|RB made|VBN little|JJ difference|NN for|IN YHL|NNP (|( columns|NNS 3|CD and|CC 4|CD )|) .|. 
Subsequent|JJ analyses|NNS are|VBP adjusted|VBN only|RB for|IN age|NN ,|, race|NN ,|, and|CC former|JJ smoking|NN .|. 
As|IN mentioned|VBN above|IN ,|, the|DT group|NN with|IN BMI|NNP from|IN 18.5|CD to|TO 20|CD would|MD be|VB considered|VBN '|POS normal|JJ '|POS by|IN the|DT NHLBI|NNP guidelines|NNS ,|, but|CC had|VBD lower|JJR YOL|NNP and|CC YHL|NNP than|IN those|DT with|IN 20|CD -|: 24.9|CD in|IN all|DT comparisons|NNS .|. 
For|IN this|DT reason|NN ,|, and|CC to|TO increase|VB sample|NN size|NN for|IN those|DT with|IN low|JJ BMI|NNP ,|, we|PRP combined|VBD the|DT two|CD lower|JJR categories|NNS ,|, defining|VBG underweight|NN as|IN a|DT BMI|NNP under|IN 20|CD .|. 

Figure|NN 1|CD is|VBZ a|DT plot|NN of|IN adjusted|VBN YOL|NNP and|CC YHL|NNP by|IN sex|NN and|CC BMI|NNP .|. 
For|IN each|DT BMI|NNP category|NN the|DT mean|NN and|CC its|PRP$ 95|CD %|NN confidence|NN interval|NN are|VBP plotted|VBD .|. 
Categories|NNP whose|WP$ confidence|NN intervals|NNS do|VBP not|RB overlap|VB ,|, or|CC overlap|VB only|RB slightly|RB ,|, are|VBP significantly|RB different|JJ .|. 
The|DT bars|NNS are|VBP slightly|RB offset|VBN to|TO permit|VB all|DT error|NN bars|NNS to|TO be|VB seen|VBN .|. 

YOL|NNP for|IN women|NNS (|( the|DT uppermost|NN curve|NN on|IN Figure|NN 1|CD )|) averaged|VBD about|IN 6.5|CD out|IN of|IN 7|CD years|NNS ,|, and|CC showed|VBD no|DT evident|JJ association|NN between|IN BMI|NNP and|CC YOL|NNP for|IN BMI|NNP above|IN 20|CD .|. 
Underweight|NNP women|NNS averaged|VBD about|IN .25|CD fewer|JJR YOL|NNP than|IN other|JJ women|NNS (|( p|NN <|NN .05|CD compared|VBN with|IN normal|JJ group|NN )|) .|. 
Underweight|NNP men|NNS also|RB had|VBD lower|JJR YOL|NNP ,|, but|CC this|DT group|NN was|VBD not|RB significantly|RB different|JJ from|IN the|DT normal|JJ group|NN ,|, in|IN part|NN because|IN of|IN low|JJ sample|NN size|NN .|. 
Men|NN classified|VBN as|IN normal|JJ ,|, overweight|NN or|CC obese|NN all|DT had|VBD about|IN the|DT same|JJ YOL|NNP .|. 

The|DT lowermost|NN two|CD lines|NNS in|IN Figure|NN 1|CD show|NN mean|NN YHL|NNP for|IN women|NNS and|CC men|NNS .|. 
Women|NNP who|WP were|VBD normal|JJ or|CC overweight|NN averaged|VBD about|IN 4.9|CD YHL|NNP .|. 
The|DT YHL|NNP for|IN underweight|NN or|CC obese|NN women|NNS was|VBD about|IN 4.5|CD years|NNS ,|, which|WDT was|VBD significantly|RB lower|JJR than|IN the|DT normal|JJ group|NN .|. 
The|DT relationship|NN of|IN BMI|NNP to|TO YHL|NNP for|IN men|NNS is|VBZ similar|JJ ,|, but|CC differences|NNS among|IN BMI|NNP groups|NNS were|VBD not|RB statistically|RB significant|JJ .|. 
YHL|NNP was|VBD significantly|RB higher|JJR for|IN women|NNS than|IN for|IN men|NNS in|IN the|DT normal|JJ and|CC overweight|NN groups|NNS ,|, but|CC the|DT sexes|NNS had|VBD similar|JJ YHL|NNP in|IN the|DT underweight|NN and|CC obese|NN groups|NNS .|. 

We|PRP next|JJ present|VB the|DT effect|NN size|NN for|IN comparing|VBG each|DT group|NN to|TO the|DT normal|JJ BMI|NNP group|NN .|. 
The|DT effect|NN sizes|NNS are|VBP shown|VBN in|IN Table|NNP 3|CD ,|, with|IN the|DT significance|NN results|NNS of|IN the|DT associated|VBN t-tests|JJ for|IN the|DT differences|NNS in|IN means|NNS of|IN the|DT two|CD groups|NNS being|VBG compared|VBN .|. 
For|IN example|NN ,|, underweight|NN women|NNS averaged|VBD 4.50|CD YHL|NNP compared|VBD to|TO 4.92|CD for|IN normal|JJ women|NNS ,|, and|CC the|DT common|JJ standard|JJ deviation|NN was|VBD 1.44|CD .|. 
The|DT effect|NN size|NN is|VBZ thus|RB (|( 4.92|CD -|: 4.50|CD )|) /|NN 1.44|CD =|SYM .29|CD .|. 
The|DT two|CD groups|NNS had|VBD significantly|RB different|JJ YHL|NNP ,|, implying|VBG that|IN the|DT effect|NN size|NN is|VBZ also|RB significantly|RB greater|JJR than|IN zero|CD .|. 
A|DT clinical|JJ trial|NN of|IN a|DT treatment|NN to|TO help|VB underweight|NN women|NNS achieve|VBP normal|JJ weight|NN (|( presumably|RB by|IN addressing|VBG the|DT underlying|VBG cause|NN )|) could|MD be|VB expected|VBN to|TO have|VB 80|CD %|NN power|NN with|IN N|NNP =|SYM (|( 1.96|CD +|NN .84|CD )|) 2/.29|CD 2|CD =|SYM about|IN 93|CD women|NNS per|IN treatment|NN arm|NN ,|, if|IN 7|CD -|: year|NN YHL|NNP were|VBD the|DT outcome|NN measure|NN .|. 

The|DT biggest|JJS effect|NN sizes|NNS are|VBP in|IN the|DT first|JJ row|NN ,|, comparing|VBG underweight|NN to|TO normal|JJ .|. 
YHL|NNP and|CC YOL|NNP have|VBP similar|JJ effect|NN sizes|NNS for|IN women|NNS ,|, and|CC are|VBP significantly|RB different|JJ from|IN zero|CD .|. 
The|DT effect|NN sizes|NNS are|VBP not|RB significantly|RB different|JJ from|IN zero|CD for|IN men|NNS ,|, in|IN part|NN because|IN there|EX were|VBD only|RB 42|CD men|NNS in|IN the|DT underweight|NN category|NN .|. 
The|DT effect|NN size|NN comparing|VBG overweight|NN to|TO normal|RB yielded|VBD small|JJ ,|, non-significant|JJ effect|NN sizes|NNS ,|, with|IN inconsistent|JJ signs|NNS ,|, suggesting|VBG extremely|RB large|JJ sample|NN sizes|NNS would|MD be|VB needed|VBN .|. 
For|IN comparing|VBG obese|NN to|TO normal|JJ ,|, only|RB YHL|NNP for|IN women|NNS showed|VBD a|DT large|JJ and|CC significant|JJ effect|NN size|NN .|. 
Thus|RB ,|, an|DT intervention|NN to|TO improve|VB the|DT health|NN of|IN underweight|NN women|NNS to|TO that|DT of|IN their|PRP$ normal|JJ weight|NN peers|NNS could|MD be|VB performed|VBN using|VBG either|CC YHL|NNP or|CC YOL|NNP as|IN the|DT outcome|NN variable|JJ .|. 
Trials|NNP to|TO make|VB obese|NN women|NNS comparable|JJ to|TO normal|JJ women|NNS could|MD be|VB evaluated|VBN using|VBG YHL|NNP ,|, but|CC not|RB YOL|NNP .|. 
Trials|NNP to|TO improve|VB the|DT health|NN of|IN the|DT other|JJ groups|NNS to|TO that|DT of|IN the|DT normals|NNS would|MD probably|RB be|VB fruitless|JJ since|IN there|EX is|VBZ no|DT evidence|NN that|IN being|VBG overweight|NN (|( for|IN men|NNS or|CC women|NNS )|) or|CC obese|NN (|( for|IN men|NNS )|) affects|VBZ YOL|NNP or|CC YHL|NNP .|. 

As|IN mentioned|VBN above|IN ,|, we|PRP repeated|VBD these|DT analyses|NNS excluding|VBG the|DT persons|NNS with|IN partially|RB estimated|VBN data|NNS ,|, and|CC using|VBG two|CD different|JJ ways|NNS of|IN coding|VBG YHL|NNP .|. 
The|DT only|RB substantive|JJ change|NN was|VBD that|IN some|DT of|IN the|DT differences|NNS between|IN blacks|NNS and|CC whites|NNS shown|VBN in|IN Table|NNP 1|CD were|VBD no|RB longer|RB statistically|RB significant|JJ ,|, due|JJ to|TO a|DT smaller|JJR sample|NN size|NN .|. 

Discussion|NNP Optimal|NNP weight|NN and|CC overweight|NN Recent|JJ studies|NNS have|VBP defined|VBN obesity|NN without|IN reference|NN to|TO age|NN [|NN 6|CD 13|CD 30|CD ]|NN .|. 
Andres|NNP et|CC al|NN proposed|VBD a|DT desirable|JJ BMI|NNP of|IN 24|CD -|: 30|CD for|IN persons|NNS aged|VBN 60|CD to|TO 69|CD [|NN 8|CD ]|NN .|. 
Allison|NNP et|CC al|NN [|NN 31|CD ]|NN proposed|VBD 27|CD -|: 30|CD for|IN older|JJR men|NNS and|CC 30|CD -|: 35|CD for|IN older|JJR women|NNS .|. 
In|IN Figure|NN 1|CD ,|, the|DT overweight|NN (|( as|IN opposed|VBN to|TO the|DT obese|NN )|) are|VBP no|DT different|JJ from|IN those|DT of|IN normal|JJ weight|NN ,|, suggesting|VBG that|IN these|DT two|CD categories|NNS could|MD be|VB combined|VBN for|IN older|JJR adults|NNS .|. 
Since|IN future|JJ improvements|NNS in|IN life|NN expectancy|NN may|MD be|VB limited|JJ [|NN 32|CD ]|NN ,|, the|DT greatest|JJS advances|NNS may|MD be|VB made|VBN by|IN improving|VBG people|NNS 's|POS YHL|NNP .|. 
This|DT suggests|VBZ that|IN the|DT development|NN of|IN future|JJ guidelines|NNS should|MD take|VB YHL|NNP or|CC other|JJ measures|NNS of|IN quality|NN of|IN life|NN into|IN account|NN .|. 

Implications|NNP for|IN clinical|JJ trials|NNS Based|VBN on|IN these|DT findings|NNS ,|, trials|NNS to|TO address|VB obesity|NN in|IN older|JJR women|NNS could|MD be|VB efficient|JJ if|IN YHL|NNP (|( but|CC not|RB YOL|NNP )|) was|VBD the|DT outcome|NN measure|NN .|. 
That|DT is|VBZ ,|, women|NNS who|WP changed|VBD from|IN being|VBG obese|NN to|TO being|VBG normal|JJ would|MD likely|JJ show|VB changes|NNS in|IN YHL|NNP ,|, but|CC not|RB YOL|NNP .|. 
Clinical|NNP trials|NNS of|IN weight|NN modification|NN interventions|NNS for|IN older|JJR adults|NNS who|WP were|VBD merely|RB overweight|NN would|MD appear|VB to|TO be|VB fruitless|JJ since|IN the|DT interventions|NNS would|MD probably|RB not|RB have|VB a|DT direct|JJ effect|NN on|IN either|DT YOL|NNP or|CC YHL|NNP .|. 

Weight|NNP or|CC weight|NN change|NN are|VBP sometimes|RB used|VBN as|IN the|DT outcome|NN in|IN evaluations|NNS of|IN interventions|NNS such|JJ as|IN diet|NN or|CC exercise|NN programs|NNS .|. 
The|DT fact|NN that|DT weight|NN is|VBZ not|RB associated|VBN in|IN a|DT consistent|JJ way|NN with|IN health|NN suggests|VBZ that|IN such|JJ evaluations|NNS should|MD be|VB considered|VBN critically|RB when|WRB older|JJR adults|NNS are|VBP the|DT subjects|NNS .|. 
This|DT is|VBZ particularly|RB important|JJ in|IN the|DT light|NN of|IN recent|JJ findings|NNS ,|, which|WDT found|VBD that|WDT interventions|NNS such|JJ as|IN weight-loss|JJ drugs|NNS may|MD be|VB harmful|JJ [|NN 33|CD 34|CD ]|NN .|. 
For|IN older|JJR adults|NNS ,|, the|DT risks|NNS associated|VBN with|IN higher|JJR weight|NN are|VBP especially|RB unclear|JJ ,|, and|CC the|DT optimal|NN outcome|NN for|IN a|DT trial|NN of|IN weight|NN loss|NN in|IN older|JJR adults|NNS requires|VBZ specific|JJ attention|NN to|TO improved|VBN health|NN and|CC mortality|NN .|. 

Interestingly|RB ,|, the|DT strongest|JJS health|NN relationships|NNS were|VBD found|VBN for|IN underweight|NN older|JJR adults|NNS .|. 
Clinical|NNP trials|NNS whose|WP$ objective|NN was|VBD to|TO make|VB the|DT underweight|NN as|RB healthy|JJ as|IN their|PRP$ normal-weight|JJ peers|NNS (|( presumably|RB by|IN addressing|VBG the|DT underlying|VBG conditions|NNS that|WDT caused|VBD the|DT low|JJ weight|NN )|) could|MD be|VB performed|VBN efficiently|RB using|VBG either|CC YOL|NNP or|CC YHL|NNP as|IN the|DT outcome|NN measure|NN .|. 
Both|DT YOL|NNP and|CC YHL|NNP would|MD be|VB clinically|RB significant|JJ in|IN this|DT patient|NN group|NN .|. 

Potential|JJ limitations|NNS CHS|NNP participants|NNS were|VBD somewhat|RB healthier|JJR than|IN the|DT average|JJ older|JJR adult|NN ;|: however|RB ,|, adjustment|NN for|IN detailed|JJ covariates|NNS made|VBD little|JJ difference|NN in|IN the|DT findings|NNS .|. 
We|PRP estimated|VBD the|DT last|JJ four|CD years|NNS of|IN health|NN data|NNS for|IN about|IN 10|CD %|NN of|IN the|DT sample|NN ,|, but|CC results|NNS with|IN and|CC without|IN this|DT group|NN were|VBD similar|JJ .|. 
Analysis|NNP of|IN mean|NN YOL|NNP instead|RB of|IN the|DT more|RBR traditional|JJ survival|NN analysis|NN survival|NN analysis|NN was|VBD appropriate|JJ here|RB ,|, since|IN virtually|RB no|DT persons|NNS were|VBD lost|VBN to|TO follow-up|JJ .|. 
Biases|NNP caused|VBN by|IN over-adjustment|JJ are|VBP probably|RB not|RB large|JJ ,|, since|IN the|DT findings|NNS were|VBD not|RB sensitive|JJ to|TO the|DT number|NN of|IN variables|NNS adjusted|VBN for|IN .|. 

These|DT results|NNS are|VBP for|IN a|DT 7|CD -|: year|NN follow-up|JJ .|. 
The|DT relative|JJ superiority|NN of|IN YHL|NNP to|TO YOL|NNP would|MD probably|RB hold|VB in|IN trials|NNS with|IN shorter|JJR follow-up|JJ .|. 
The|DT effect|NN sizes|NNS in|IN Table|NNP 3|CD might|MD also|RB be|VB appropriate|JJ in|IN shorter|JJR trials|NNS ,|, since|IN lengthy|JJ trials|NNS often|RB add|VBP little|JJ information|NN [|NN 10|CD ]|NN .|. 

EVGFP|NNP ,|, on|IN which|WDT YHL|NNP was|VBD based|VBN ,|, might|MD have|VB missed|VBN some|DT effects|NNS of|IN obesity|NN on|IN risk|NN factors|NNS for|IN future|JJ health|NN .|. 
A|DT person|NN who|WP is|VBZ depressed|VBN because|IN of|IN a|DT poor|JJ self-image|JJ related|VBN to|TO obesity|NN or|CC who|WP has|VBZ osteo-arthritis|JJ related|VBN to|TO obesity|NN and|CC limits|NNS to|TO activities|NNS to|TO successfully|RB avoid|VB pain|NN would|MD surely|RB have|VB worse|JJR EVGFP|NNP than|IN others|NNS ,|, based|VBN on|IN results|NNS from|IN many|JJ studies|NNS .|. 
However|RB ,|, health|NN measures|NNS designed|VBN specifically|RB to|TO measure|VB those|DT conditions|NNS might|MD be|VB more|RBR sensitive|JJ to|TO change|VB in|IN weight|NN than|IN EVGFP|NNP .|. 
If|IN YHL|NNP were|VBD based|VBN on|IN such|JJ measures|NNS ,|, the|DT superiority|NN of|IN YHL|NNP to|TO YOL|NNP would|MD likely|RB be|VB even|RB greater|JJR than|IN that|DT shown|VBN here|RB .|. 
These|DT more|RBR sensitive|JJ measures|NNS might|MD also|RB have|VB detected|VBN differences|NNS between|IN the|DT overweight|NN and|CC normal|JJ weight|NN persons|NNS ,|, but|CC we|PRP think|VBP this|DT is|VBZ unlikely|RB given|VBN the|DT absence|NN of|IN any|DT differences|NNS in|IN EVGFP|NNP .|. 

Conclusion|NNP Recommendations|NNP for|IN desirable|JJ weight|NN have|VBP been|VBN criticized|VBN for|IN emphasizing|VBG mortality|NN rather|RB than|IN health|NN .|. 
We|PRP found|VBD associations|NNS between|IN YHL|NNP and|CC obesity|NN that|WDT were|VBD not|RB present|JJ in|IN the|DT mortality|NN analysis|NN ,|, suggesting|VBG that|IN YHL|NNP may|MD be|VB a|DT more|RBR sensitive|JJ measure|NN of|IN the|DT burden|NN of|IN obesity|NN in|IN older|JJR adults|NNS ,|, especially|RB for|IN women|NNS .|. 
Future|JJ efforts|NNS to|TO determine|VB desirable|JJ weight|NN guidelines|NNS should|MD include|VB measures|NNS of|IN YHL|NNP .|. 
Using|VBG either|CC YOL|NNP or|CC YHL|NNP ,|, however|RB ,|, we|PRP found|VBD no|DT excess|JJ risk|NN for|IN older|JJR adults|NNS who|WP would|MD be|VB classified|VBN as|IN '|POS overweight|NN '|'' by|IN the|DT NHLBI|NNP guidelines|NNS .|. 
This|DT suggests|VBZ using|VBG YHL|NNP as|IN the|DT outcome|NN measure|NN in|IN clinical|JJ trials|NNS involving|VBG obese|NN or|CC underweight|NN older|JJR adults|NNS ,|, and|CC discouraging|VBG trials|NNS that|WDT address|VBP older|JJR adults|NNS who|WP are|VBP merely|RB overweight|NN .|. 

Competing|VBG interests|NNS None|NN declared|VBD 
Abbreviations|NNP BMI|NNP Body|NNP mass|NN index|NN 
CESD|NNP Center|NNP for|IN Epidemiologic|NNP Studies|NNPS Depression|NNP Scale|NNP 
CHS|NNP Cardiovascular|NNP Health|NNP Study|NNP 
EVGFP|NNP Is|VBZ your|PRP$ health|NN excellent|JJ ,|, very|RB good|JJ ,|, good|JJ ,|, fair|JJ or|CC poor|JJ ?|. 
QALY|NNP Quality-adjusted|NNP life|NN years|NNS 
YHL|NNP Years|NNPS of|IN healthy|JJ life|NN 
YOL|NNP Years|NNPS of|IN life|NN 
