Imprecise temporal and causal requirements

2016-Mar-16

Comments

Mar 16, 2016

Bruce

Following on from the call, you asked for examples of requirements which seemed to include a loosely-specified temporal or causal association.

For causal associations, I encountered the "Vasculitis associated with probable etiology" category within the 2012 CHCC vasculitis nomenclature (1). We added some of these to the UK extension (pending move) last year, framing them more firmly, for example as 'Vasculitis co-occurrent and due to Hepatitis B virus infection (disorder)' (PT 'Hepatitis B virus-associated vasculitis'). As you can see the way they are presented in the CHCC document indicates a desire to 'downgrade' any causal association.

For temporal associations, we added a family of 'Injury following substance use' concepts last year (one example being 'Injury following crystal methamphetamine use'). The way these are worded (and the way they might therefore be modelled) puts them firmly in an 'after' pattern, but if so, 'after' what? If 'after' refers to a 'commencement of substance use' event then phrases of this form are valid for injuries incurred at any point from intake to completely clearance (but are probably intended to mean one of the earlier states). By contrast if 'after substance use' is considered comparable to the way we think about 'after a procedure' or 'post infection' (ended/resolved) then strictly they only apply to 'late injuries' after the 'substance use' has ended. For info 'Injury whilst under the influence of...' was not felt to be a suitable alternative phrasing, as was any reference to causation. The ability to soften the semantics of 'after' therefore to include 'overlaps with' for this scenario is appealing, but whether it's feasible is a different matter.

(1) http://onlinelibrary.wiley.com/doi/10.1002/art.37715/epdf

Bruce Goldberg

Mar 16, 2016

Hi Ed. Thank you for providing these examples. I have been playing around with an OWL representation of temporal relationships that expands upon the associated with role hierarchy to include during and some degree of overlap between during and after with the intent to define surgical complications. It has occurred to me that overlap between during and after may not necessarily represent the entire time interval during which a complication occurs but rather a period that may include during or after or the entire period spanning during and after. Thus creating during AND/OR after as a parent of both during and of after and defining perioperative complication as a complication during and/or after surgery, results in preoperative and postoperative complications as subtypes of perioperative complications which I think makes sense. This approach might also work for other complications such as you mention in which e.g. an injury might occur during, after or during and continuing after an event has resolved.

On another note, yesterday I completed my recertification as a clinical investigator and one of the modules I needed to complete dealt with causation as related to adverse events. I have included an excerpt for interest sake and not to suggest we need to move in this direction.

In 2005, the Council for International Organizations of Medical Sciences (CIOMS VI) working group recommended a binary causality assessment of Related or Not Related (Management of Safety Information from Clinical Trials, Report of CIOMS Working Group VI). In addition to this recommendation, research sponsors could also provide categories for evaluating the relationship between the AE and the test product.

Common categories include:

Definitely related: There is a certainty that the event is related to the investigational product.
Probably related: There is high likelihood that the event is related to the investigational product.
Possibly related: There is a likelihood that the investigational product is the cause of the event, but other causes cannot be ruled out.
Unlikely to be related: It is not likely that the event is related to the investigational product, and other more likely causes are present.
Unrelated: Evidence exists that the event is related to something other than the investigational product.

Table 4 illustrates how the National Institute of Allergy and Infectious Disease (NIAID) uses the principles discussed in Sections on "Determining Causality" and "Categories of Causality," combining and evaluating multiple factors to define categories of causality.

Table 4. Relatedness of AEs to an Intervention (Product)
Definite (must have all 4)	Has a reasonable temporal relationship to the intervention Could not have readily been produced by the subject's clinical state or have been due to environmental or other interventions Follows a known pattern of response to intervention Disappears or decreases with reduction in dose or cessation of intervention and recurs with re-exposure
Probable (must have 3)	Has a reasonable temporal relationship to the intervention Could not have readily been produced by the subject's clinical state or have been due to environmental or other interventions Follows a known pattern of response to intervention Disappears or decreases with reduction in dose or cessation of intervention
Possible (must have 2)	Has a reasonable temporal relationship to the intervention Could not have readily been produced by the subject's clinical state Could not readily have been due to environmental or other interventions Follows a known pattern of response to intervention
Unlikely (must have 2)	Does not have a temporal relationship to the intervention Could readily have been produced by the subject's clinical state Could have been due to environmental or other interventions Does not follow a known pattern of response to intervention Does not reappear or worsen with reintroduction of intervention

Ed Cheetham

Mar 21, 2016

Thanks

For completeness here is a snippet from an email exchange between you, me and Yong from last July touching on the same subject, where we came to a similar conclusion on the utility of various temporal relations. That said I'm still uncomfortable about the temporally disjunctive class (during and/or after) in that it effectively represents 'not co-occurrent', which still leaves considerable uncertainty as to whether the Y of 'X during and/or after Y' is included in the relevant life phase:

From July 17 2015 "...Based on our discussions on Monday, of the three possible new additions in white (see attached figure) I’d be inclined to campaign first for ‘after or overlaps with’ – since this then covers all the Allen’s time categories except co-occurrence for which we are recommending multiple is_a’s ({p, m, o, s} vs. {F, D, e, S}).

I agree in the attraction of adding ‘due_to and after’, but this can already be represented by the combination of existing roles. I’m not sure there is a clinical use for ‘overlaps with’ on its own.

As Bruce hinted on the call, ‘after or overlaps with’ is perilously close to ‘associated with’ in its vagueness, but is perhaps it is a *bit* crisper in its semantics. It would support (objectively, given the Allen’s analysis) a time-only mechanism for associating the two components of an X + Y construct where co-occurrence is not necessarily implied, and it would be silent on causation..."

Kind regards

Bruce Goldberg

Mar 21, 2016

Hi Ed. Thanks for refreshing my memory on this. My rationale for considering a disjunctive (during and/or after) class is based on a specific use case for representing perioperative complications. My understandings about these are that they comprise intraoperative complications, postoperative complications or both (a complication that begins during surgery and persists throughout the postoperative period). It is thus just this uncertainty I was trying to capture. Perhaps after or overlaps with would work as well but in order for intraoperative and postoperative complications to be subsumed by perioperative complications this role would need to be a parent of both during and of after which is reversed in the diagram above.

Bruce