Question
Question Posted 09/24/26:
Dear experts,
I am writing to ask for clarification regarding the PTADDX variable in the diagnosis information. According to the data dictionary, PTADDX represents the year of Alzheimer’s disease diagnosis. However, we have noticed that this information is missing for a considerable number of participants diagnosed with AD. Our first question is whether, for AD participants with missing PTADDX values, it would be appropriate to use the screening date as an approximation of the diagnosis date, or whether this should be avoided. We need an estimate of the year of diagnosis in order to calculate disease duration from the start. We are also considering whether PTCOGBEG would be more appropriate for this purpose. According to the dictionary, PTCOGBEG represents the best estimate of the year of onset of cognitive symptoms. Could you please advise which of these variables (PTADDX or PTCOGBEG) would be more appropriate for calculating disease duration in AD participants from the onset? In particular, we would appreciate any guidance on whether disease duration should preferably be defined from the estimated onset of symptoms (PTCOGBEG) or from the clinical diagnosis (PTADDX), and how you would recommend handling missing PTADDX values.
Thank you very much for your help and clarification,
Dear experts,
I am writing to ask for clarification regarding the PTADDX variable in the diagnosis information. According to the data dictionary, PTADDX represents the year of Alzheimer’s disease diagnosis. However, we have noticed that this information is missing for a considerable number of participants diagnosed with AD. Our first question is whether, for AD participants with missing PTADDX values, it would be appropriate to use the screening date as an approximation of the diagnosis date, or whether this should be avoided. We need an estimate of the year of diagnosis in order to calculate disease duration from the start. We are also considering whether PTCOGBEG would be more appropriate for this purpose. According to the dictionary, PTCOGBEG represents the best estimate of the year of onset of cognitive symptoms. Could you please advise which of these variables (PTADDX or PTCOGBEG) would be more appropriate for calculating disease duration in AD participants from the onset? In particular, we would appreciate any guidance on whether disease duration should preferably be defined from the estimated onset of symptoms (PTCOGBEG) or from the clinical diagnosis (PTADDX), and how you would recommend handling missing PTADDX values.
Thank you very much for your help and clarification,
Response posted 09/24/26 by Adam Diaz:
That field does not exist in the diagnostic information table; the table you'll want to use as aprimary source of diagnostic information is listed on the IDA as 'diagnostic summary', and the filename isDXSUM. The PTADDX and PTCOGBEG fields are located in the participant demographics table. The former table should be considered the primary source of information on diagnostic classification, and more information on it can be found in the ADNI data user guide located here (https://adni.loni.usc.edu/help-faqs/adni-data-user-guide/). The diagnoses in the DXSUM table are clinically assessed in accordance with ADNI diagnostic criteria, while the PTADDX and PTCOGBEG fields in the demographics table are self-reported as part of the participant demographics questionnaire, with some caveats. The PTADDX question is administered only to participants in the dementia cohort (defined per the DXSUM table), while PTCOGBEG is administered to participants in either the MCI or dementia cohorts. They have some utility in capturing the duration of symptoms prior to participant entry into the study, but they are both self-reported, conditionally administered, and probably not the most reliable.
Response posted 09/25/26 by Katherine Robles:
Thank you very much for your response and for the clarification. It's right, I was referring to the demographics data rather than the diagnostic information. My apologies for the confusion.
Given that PTADDX and PTCOGBEG are self-reported and may therefore not provide the most reliable estimate, is there any other variable in ADNI database that could be used to determine or approximate disease onset? For example, is this information available in any of the clinical or physiological assessment tables? Alternatively, for participants classified as having dementia in DXSUM at screening, would it be appropriate to use the screening EXAMDATE as an approximation of the year of diagnosis?
Our main objective is to obtain the most reliable estimate possible of disease duration for participants with AD, so any recommendation on how this could best be derived from the available ADNI data would be greatly appreciated.
Thank you again for your help.
Given that PTADDX and PTCOGBEG are self-reported and may therefore not provide the most reliable estimate, is there any other variable in ADNI database that could be used to determine or approximate disease onset? For example, is this information available in any of the clinical or physiological assessment tables? Alternatively, for participants classified as having dementia in DXSUM at screening, would it be appropriate to use the screening EXAMDATE as an approximation of the year of diagnosis?
Our main objective is to obtain the most reliable estimate possible of disease duration for participants with AD, so any recommendation on how this could best be derived from the available ADNI data would be greatly appreciated.
Thank you again for your help.
Response posted 09/25/26 by Adam Diaz:
Considering examdates where a DX of impairment was assigned as censored (left-censored for diagnoses at screening, and interval censored in the case of progression from a CU diagnosis) is the most principled approach that comes to mind, but this may or may not be amenable to your specific research question/modeling approach. Any other fields that record something similar (e.g. entries in the medical history/initial health assessment tables) will suffer from the same self-report inconsistency.
