Details
-
Change Request
-
Resolution: Not Persuasive
-
Medium
-
FHIR Core (FHIR)
-
STU3
-
Modeling & Methodology
-
Datatypes
-
-
Grahame Grieve / Ron Parker: 3-0-0
Description
An exclusive range data representation is needed in FHIR to address accurately representing acceptance criteria for analytic tests. "…analytical methods should be scientifically sound (e.g., specific, sensitive, and accurate) and provide results that are reliable.", FDA Process Validation Guidance (2011) +. + This specificity, sensitivity and accuracy is accomplished via statistics. When statistics are supported by large sample sizes and normal distributions the result is often acceptance criteria that can be expressed with inclusive numeric ranges which are currently captured in Range where it states "The low and the high values are inclusive, and are assumed to have arbitrarily high precision; e.g. the range 1.5 to 2.5 includes 1.50, and 2.50 but not 1.49 or 2.51."
However, not all data available for setting acceptance criteria have this "high precision" when the specifications are being developed. Statistics are hampered by small sample sets and outliers, or both. When faced with data that does not yield that perfect 3-sigma, other statistical methods are employed. Under the assumption that preproduction lots may not fully represent future production values, outliers are removed to obtain a normal distribution or there are no outliers, yet there is no normal distribution. The resulting specification limits are burdened with uncertainty. The acceptance criteria can be set that can curb that uncertainty by stating that if those outlier values are seen again, well, that it certainly is a failure. Those limits are often exclusionary. So along with a mean, there will be limits like less than 97 or greater than 93. Also, exponential distributions do to the magnitude at the tail, will often have a one limit that is exclusive.
Regardless of the source, there are exclusive limits in specification acceptance criteria. The concept of exclusive ranges is not novel or unique to PQ/CMC. There is a common math notion with brackets for inclusive and parenthesis for exclusion for indicating ranges where one or both of the starting and ending values are exclusive, meaning that the starting/ending value is not included in the range of values. This is an obstacle for defining profiles for PQ/CMC. The range data type in PlanDefinition is inclusive and the Observation.referenceRange is also inclusive. The USP rounding rules documents are attached for clarification of why the exclusive range is required for PQ/CMC reporting.
My name if not in the list of names. Catherine Hosage Norman chn@module3solutions.com