Present: @GracielaMuniz, @annierobi, @emielhoogendijk , @jochinho , Judith

idea in big strokes

We inherit the methodology from iasla-2016-amsterdam and iasla-2017-amsterdam and focus on measures of physical performance: hand grip and speed gait

Studies

• LASA - (NL)
• RUSH-MAP - (USA)
• OCTO-Twin (SW)
• H70 (SW)

maybe:

• Lothian
• Newcastle85+
  However, we are not certain that adding more studies will make the paper(s) better.

Notes

• "The trick is how to describe the results" (G.Muniz)
• The gait measure needs clarification: Rapid vs Normal
• Currently, early January is what we are thinking about the in-person meeting. There will be an IALSA meeting in January in Chicago, so it might be easier to fly the European partners over.

What writer needs

Looking back at 2016 and 2017, what does the writer need to organize the manuscript? We started making bullets:

• study Description
• attrition
• sample size (over waves)
• Respondents vs Observations : make sure this is clearly spelled out
• number of deaths
• Measurement: be very specific how each measurement is coded
• Describe the temporal structure of the study: measurement waves
• How each parameter of the mode is constrained
• Whether misclassification was applied and how

Action items

• next meeting is scheduled for Monday, Nov 20, 4 pm. Amsterdam time (2017-11-20-16:00 GMT+1)
  Before meeting next time, we have agreed to

@emielhoogendijk

• send out a list of variables that will be used in the project
• send out a cutoff scores for gait and grip that would operationally define physical impairment

@andkov

• work with the What writer needs list and prepare a draft report using MAP data from ialsa-2017-amsterdam with intent to automate some of the output

@GracielaMuniz

• clarify the date for which the IALSA group meeting will be held in Chicago in Jan of 2018

by @emielhoogendijk

Dear all,

Some information for the TC tomorrow:
I have been reading the literature about cut-offs with regard to slow gait speed and weak grip strength. Most articles from the frailty field use the frailty criteria of Fried et al (2001), which includes gait speed and grip strength. Everybody copies their cut-offs, which is not completely logic as they were based on the lowest quintiles from one database, the Cardiovascular Health Study (height/sex specific for gait speed, and BMI/sex specific for grip strength, see the article in the attachment). We could follow the approach of Fried (either using their cut-offs, or derive cut-offs from lowest quintiles in our datasets) or we could go for other established measures, like a general gait speed cut-off (not taking into account sex and height).

For gait speed we can choose two approaches:

1. Use the <0.6 m/sec cut-off (see the JAMA papers of Studenski and Cummings), without any specification for sex and height.
2. Use sex and height specific cut-offs based on quintiles (derived from our own datasets, or by copying the cut-offs from Fried).

For grip strength there are no general cut-offs, as far as I know:

1. We have to decide if we use the cut-offs from Fried or the BMI & sex specific quintiles in our datasets. In many sarcopenia (see article Cruz Jentoft et al. in attachment) and frailty papers, the Fried cut-offs are being used.

I don`t know if we combine gait speed and grip strength in one paper? I would prefer to focus on gait speed first (as there is now a lot of literature about the slow gait speed as early signal of motoric problems related to starting dementia, see for example the article of Del Campo in the attachment – and also in geriatrics slow gait speed is seen as the best indicator of declining health). What would be interesting is to compare the 0.6 m/sec cut-off with the lowest quintile approach, to see whether it gives consistent results across studies. This could result in a paper that is interesting for JAGS (Journal of the American Geriatrics Society).

By @emielhoogendijk

Inherited from previous

As it is an extension of our previous analyses, all the other variables should remain the same:
· Age (time dependent – we have to discuss about the centering)
· Sex (same as before)
· Educational level (continuous, centering we have to discuss)
· State variable (same as before, based on MMSE: no cognitive impairment, mild cognitive impairment, severe cognitive impairment and death)

New to 2018

In addition to inherited variables we need:
· Gait speed in seconds (please check the distance measured in your study, as well as the protocol: normal pace or rapid pace)
· Height in cm
· Grip strength (if available) in Kg (please check which dynamometer was used and in which position the grip strength was measured)
· Body Mass Index (body weight in kilograms divided by height in meters squared).

Notes

We may discuss if we need to adjust for comorbidity (number of chronic diseases) or not. But in our previous analyses we also did not control for many covariates, so maybe it is not necessary.

I think we also don`t need income. Educational level is more important.

Cut-offs for gait speed (sex/height specific or not) is still to be determined.

Some points for discussion:

(1). Convention for model and submodels nomenclature

(2). Does everyone have physical activity?

(3). Do we separate analyses by gender?

(4). What are the specific instruction for age*gender specific cutoffs for gait?

(5). What should be the method for converting time-variant measures to time-invariant?

• baseline? (the first non-missing observation)
• median (across observable lifespan)
• something else (personal slope? difference score?)

Here's the solution Ardo prepared for selecting those individuals who have a valid observation of gait between ages 68 and 72 (inclusive)

dta <- ds_valid %>% dplyr::select(-firstobs)
subjects <- unique(ds_valid$id) 

library(dplyr)
library(elect)
# dta <- ELECTData
# subjects <- unique(dta$id)
for(i in 1:length(subjects)){
  dta.i <- dta[dta$id==subjects[i],]
  if(nrow(dta.i)<=2){print(subjects[i])}
}

dta %>% glimpse()
# Subselect:
bound <- c(68,72)
count <- 0
for(i in 1:length(subjects)){
  select <- 0
  dta.i <- dta[dta$id==subjects[i],]
  if(dta.i$age[1]< bound[2]){
    ddta.i <- dta.i[dta.i$age>bound[1],]
    if(nrow(ddta.i)>1){
      select <- 1; print(i)
      select <- 1 ; print (i)
      firstobs <- rep(0, nrow(ddta.i))
      firstobs[1] <- 1
      ddta.i <- cbind(ddta.i, firstobs = firstobs)
    }
  }
  if(select==1 & count==0){
    ddta <- ddta.i
    count <- count +1
  }
  if(select==1 & count>0){
    ddta <- rbind(ddta,ddta.i)
    count <- count+1
  }
}

hist(ddta$age[ddta$firstobs == 1])