Phases and Steps of the analytic workflow

[andkov]

Amsterdam - minumum reporting requirements and standard workflow

Stages:

• 0-ellis-island
• 1-encode-multistate-mmse
• 2-estimate-models
• 3-tabulate-results
• 4-report-results

Each stage can be broken down into a series of step.

Stage 0 - Ellis Island

• [ ] Step 1 - Import raw data
• [ ] Step 2 - Create a starter for the meta-data object (export names and labels to external .csv file)
• [ ] Step 3 - Augment meta-data to provide useful labels, descriptors, and selectors
• [ ] Step 4 - Apply meta-data to transform the raw data

  Stage 1 - Encode States

• [ ] Step 1 - Describe the data before encoding multisates
  • Print data for a handful of ids in long format for variables to be used in multistate modeling
  • Examine the effect of attrition(Count alive and dead subjects at each time point)
  • Graph observed trajectories of MMSE across time and age (variable used to encode states)
• [ ] Step 2 - Encode missing states: (interval missing (-1) and right-censored (-2). (create ds_miss)
• [ ] Step 3 - Encode multi-states (create ds_ms)
• [ ] Step 4 - Inspect data for incorrect encoding. Compare data before (ds_miss) and after (ds_ms) encoding of multi-state
• [ ] Step 5 - Produce the matrix of observed transitions

  Stage 2 - Estimate models

• [ ] Step 1 - Finallize data for modeling ( create ds_clean, the object to be passed to msm() call)
  • Remove subjects with missing age
• [ ] Step 2 - Describe age composition of the resultant sample
  • Distribution of age at baseline
  • Distribution of ages in data
  • Distribution of intervals between consecitive observation
• [ ] Step 3 - Specify the model (create qnames and Q objects) and center covariates
• [ ] Step 4 - Estimate multistate models using msm() call
• [ ] Step 5 - Compute Life Expectancies (LE) using elect() call
• [ ] Step 6 - Save models to disk to be used in reports