A cross-sectional study was conducted on community-dwelling participants aged 65 years or older who visited the Toulouse frailty clinic during 2011 and 2016. Each patient was referred by a physician (general practitioner, geriatrician or specialist) who had reported signs or symptoms of frailty using the Gérontopôle Frailty Screeening tool .
Patients who were referred by a physician came to the Toulouse frailty day hospital for a multidisciplinary evaluation. Socio-demographic, anthropometric, detailed medical history, functional, frailty status and disability was recorded, as well as blood sample collection.
Patients who underwent a comprehensive geriatric assessment and had a blood draw were assessed for eligibility.
Patients referred for an onco-geriatric evaluation were excluded from the study because they have an on-going inflammatory state (N = 419), as well as patients treated with erythropoietin (N = 7).
Frailty syndrome was evaluated according to the phenotype proposed by Fried et al.  based on the five criteria: unintentional weight loss, self-reported exhaustion, weakness, slow walking speed and low physical activity. Physical activity was the only adapted criterion as the Minnesota Leisure Time Questionnaire was not feasible in clinical practice. The questionnaire from the InChianti study on regular physical activity was used instead .
Weigth loss was defined as an unintentional loss of > 5 kg in the past year .
Exhaustion was present, if the participant answered often or most of the time for the question « How often in the last week did you feel that everything that you did was an effort? » used in the Center for epidemiologic studies-Depression scale .
Low physical activity was described as an absent to minimal activity level in the past year.
Slow walking speed was defined by gender and height specific cut-off values proposed by Fried over a 4 m course at usual pace.
Weakness was evaluated by hand grip strength measured by a handheld dynamometer (Jamar, Inrvington, NY) and based on gender, BMI specific cut-off values proposed by Fried . Measures were done twice and on both hands, the average of the best results were used.
Following this evaluation, participants were considered frail if they had more than 3 criteria, the others were considered non frail.
Blood tests were performed in the morning at enrollment in the frailty day clinic hospital. Samples were then sent and processed on automated instruments in the Toulouse University hospital laboratories. Hemoglobin concentration g/dl was measured using the hematology analyzer Sysmex spectrophometry using cyanmethemoglobin method.
Our study was designed to explore the relationship between frailty and hemoglobin count while controling for covariates that modify this relationship. This was done in order to see the pathways that exist in this relationship between frailty and hemoglobin concentration. Covariates likely to influence the main association tested between hemoglobin and frailty status were selected a pirori based on literature and added by order of influence. Covariates such as inherent demographics (age,sex) and health indicators were included. The other covariates were chosen based on their common association in the literature with frailty and hemoglobin concentration: kidney function, inflammation, cognition and nutritional status and socioeconomic positions [16, 22].
Two main types of covariates were distinguished:
Inflammation defined by a C-reactive protein level above 10 mg/dl . Serum levels of high-sensitivity C-reactive protein (hs-CRP) is measured by immunoturbidimetry (Roche Cobas analyzer) .
Renal function is estimated with glomerular filtration rate (GFR) calculated by using the chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation . Serum creatinine level was assessed using a Diazyme’s enzymatic method (Roche Cobas analyzer). The GFR categories were mapped to the categories retained by The Kidney Disease: Improving Global Outcomes (KDIGO) guidelines . Normal kidney function was defined as a GFR ≥ 90 ml/min per 1.73m2, mildly decrease GFR between 60 and 89 ml/min per 1.73m2, moderate to severe decrease GFR 59-30 ml/min per 1.73m2 and severe decrease GFR > 29 ml/min per 1.73m2.
The Mini nutritional assessment (MNA) was used to evaluate nutritionnal status . A MNA score ≥ 24 indicated an adequate nutritional status, a MNA score < 17 a protein-calorie malnutrition and a MNA score between 17and 23.5 a risk of malnutrition.
The Mini Mental State Examination (MMSE) as developed by Folstein was used as a surrogate for cognitive function [28, 29]. A MMSE score above 26 was considered as absence or questionable dementia, between 21 to 25 for mild, 11 to 20 as moderate and under 10 as severe dementia 
To assess the subjects’s socio-economic position, we selected proxy variables such as the level of education. Education levels are categorized using the International Standard Classification of Education 2011 . Education categories were defined as low (unschooled or primary education), medium (middle school to high school) or high (university level). We also collected living arrangements defined as either living alone or living with others (spouse, family,...).
Sample characteristics were first described. Data are reported as percentage or as mean ± standard deviation.
We tested the normal distribution for the quantitative variables using the Shapiro wilk test.
Hemoglobin concentration was entered as a continuous measure as there was a linear relationship with frailty.
Bivariate analysis were used to assess the relationship between the covariates to hemoglobin and to frailty. Significance was tested using chi-square tests for categorical variables, Wilcoxon rank sum tests and Kruskall Wallis for continuous variables as appropriate.
Associations between age and hemoglobin were measured using the Pearson correlation coefficient.
To explore whether the increased risk of being frail was associated with hemoglobin concentration reflected the presence of comorbidity rather than constituing an additional risk factor of being frail, we further adjusted for kidney function, inflammation, cognition and nutritional status traditionnaly associated with frailty and hemoglobin concentration and we finally adjusted for socioeconomic position.
Multivariate logistic regression analysis with a forward selection was used to to examine the influence of these covariates further. Series of logistic regression models were performed.
Starting with the addition in the differents models of inherent individual covariates such as demographic data (Model 1), then biological parameters: renal function (Model 2), and inflammatory parameter (Model 3) and finally variables with environmental influence such as cognitive (Model 4), nutritional (Model 5) and socio economic variables (Model 6).
In Model 7 we added in our regression model the interaction of inflammation and nutritional status with hemoglobin concentration corresponding to the full model.
We determined the respective statistical contributions of confounders in explaining the association between hemoglobin concentration and frailty by using a traditional approach to mediation.
This analysis was conducted to investigate the possible combined effects of inflammation and nutrition on hemoglobin concentration.
The analysis was performed using STATA® version 11 (Stata Corp.,College Station, TX).