![New sensor?
Y
Assume default gradient
of [x].
Base intercept on
[mean] of initial dual
calibration.
Use standard error
[how?].
N
Seek calibration based
on [certain conditions].
High enough
calibration validity
weight for inclusion
Y
Add calibration to
chart with weight
determined by
weighting algorithm;
and also use this to re-
weight prior points
N Discard calibration
Sensor calibration – conceptual framework Dan Evans, 17/12/16
Calibration validity weight
Determine this weight based on higher rate when:
• Calibration not when reading rising / falling [by
more than x mmol/min]
• Calibration in ‘normal’ range of [x-x]
• Calibration not an outlier from other points [x%
off existing line, when x points already]
• Calibration not at a time of high noise [when
latest reading off trend of last x points by x%]
Metrics for the above, and significance of the
conditions to be tested by use of large-scale user
data
Seek calibration conditions
The first of:
• When no calibration at all
• When no calibration for [x] hours
• ? When few [less than x in ‘normal
range’] data points at all / in normal
range
Metrics for the above, and significance
of the conditions to be tested by use of
large-scale user data
Weighting algorithm
Higher weighting (by x%) for calibrations which are:
• [x] hours newer
• Not in ‘bed in’ [first day] time
• Have higher calibration validity weight
Where calibration validity weight is low, allow this if few points
(which suggests that algorithm should be about relative weights).
Metrics for the above, and significance of the conditions to be tested
by use of large-scale user data
Calibration chart
Linear / curvilinear choice to be tested by
use of large-scale user data](https://image.slidesharecdn.com/calibrationframework1-161217231323/85/Calibration-framework-1-1-320.jpg)
Calibration framework 1
- 1. New sensor? Y Assume default gradient of [x]. Base intercept on [mean] of initial dual calibration. Use standard error [how?]. N Seek calibration based on [certain conditions]. High enough calibration validity weight for inclusion Y Add calibration to chart with weight determined by weighting algorithm; and also use this to re- weight prior points N Discard calibration Sensor calibration – conceptual framework Dan Evans, 17/12/16 Calibration validity weight Determine this weight based on higher rate when: • Calibration not when reading rising / falling [by more than x mmol/min] • Calibration in ‘normal’ range of [x-x] • Calibration not an outlier from other points [x% off existing line, when x points already] • Calibration not at a time of high noise [when latest reading off trend of last x points by x%] Metrics for the above, and significance of the conditions to be tested by use of large-scale user data Seek calibration conditions The first of: • When no calibration at all • When no calibration for [x] hours • ? When few [less than x in ‘normal range’] data points at all / in normal range Metrics for the above, and significance of the conditions to be tested by use of large-scale user data Weighting algorithm Higher weighting (by x%) for calibrations which are: • [x] hours newer • Not in ‘bed in’ [first day] time • Have higher calibration validity weight Where calibration validity weight is low, allow this if few points (which suggests that algorithm should be about relative weights). Metrics for the above, and significance of the conditions to be tested by use of large-scale user data Calibration chart Linear / curvilinear choice to be tested by use of large-scale user data