Hi, I recently asked a question about how the DX variable from the ADNIMERGE CSV file is computed, as I am trying to replicate it using the DXCURREN, DXCHANGE and DIAGNOSIS variables in the Diagnostic Summary CSV file, but find some discrepancies between files. In particular, some patients are not classified has having dementia using ADNIMERGE.csv, but are using DXSUM_PDXCONV_ADNIALL.csv. For example, patient with RID= 51 has DX= MCI in ADNIMERGE up to month 36 (VISCODE = m36), and then have missing values for DX for later visits (i.e. months 72 to 120). Based on DXSUM_PDXCONV_ADNIALL.csv, this patient would also have DXCURRENT= 2 (i.e., MCI) up to month 36 (VISCODE2 = m36), but would be classified as having dementia thereafter, that is DXCHANGE= 3 (i.e., 3 = Stable: Dementia to Dementia) for months 72 to 120. Similar patterns happen for 181 patients (e.g., RID= 61, 106, 112, 123, 972, 4240). Could you please confirm how the DX variable was derived in ADNIMERGE and/or confirm the rationale used when there are discrepancies?
I just downloaded the ADNIMERGE file and see diagnostic changes for recorded for several of the RIDs that you mention (I didn't check all of them, but the ones I did check had diagnostic changes). There are some missing entries in ADNIMERGE for DX, but that's partially due to visits where only MRI was done (for example, m03). RID 51 matches exactly by VISCODE with what is given in DXSUM_PDXCONV_ADNIALL.csv.
