Biometric data from Health Patch for glycemic modeling

D. Mul1,P. Funke2, M. Monteiro2, R. Vandenberg2, HJ Aanstoot1, R. Bruinsma3, HJ Veeze1

1Diabeter, centre for pediatric and adolescent diabetes care and research Rotterdam, The Netherlands; 2Aspire Ventures, Lancaster PA, USA; 3Tempo Health, Rotterdam, The Netherlands

Introduction

Predictive glycemic modeling of T1DM
patients can help guide insulin
replacement therapy. The addition of
biometric data into glycemic models
might improve the accuracy of glucose
predictions.

Materials and Methods

OBSERVATIONAL TRIAL DATA

Biometric data were collected from 20 T1DM
patients using commercially available combined
continuous glucose monitors and pumps
(Medtronic) and a patch-type, non-invasive
biometric sensor (Vital Connect HealthPatch,
figure 1).

Participants provided (figure 2)

  • continuous biometric data: heart rate, skin temperature, 3-axis accelerometry, and skin impedance
  • pump data, including, basal insulin, bolus insulin, meal-time carb estimates
  • blood glucose values by finger stick and
  • continuous blood glucose values (CGM)

MODELING

Aspire combined biometric data with finger stick
BGL readings, carbohydrate inputs, and insulin
delivery tracking to build personalized BGL
prediction models (figure 3), adapted to each
patient on a daily basis with its proprietary
Adaptive Artificial Intelligence platform (A2I).

Tempo Health’s Rhythm system has the capability to shut
itself off when it learns that is not able to reliably predict
and control the patient’s BG


Results

From the 20 patients 12 were excluded for
incomplete datasets or less than 14 days
combined data.

Eight patients included (6F), mean of most
recent HbA1c: 7%, mean age 25.9 yrs
(range 10-53)

In 7 of the 8 patients observed:
by leveraging those predictive models in a
very basic control system, compared to
the sensor readings that were actually
reached by the patient in that time period
treated by the Diabeter team:

Rhythm was able to achieve

  • 20% increase in time in
    range (70-180 mg/dl or 3.9 -
    10.0 mmol/L)
  • 9% reduction in values
    below 70 mg/dl (3.9 mmol/L)

In one patient no reliable prediction could be
obtained.

No statistically significant relationship
between frequency of fingersticks and
Rhythm results.

Conclusions

Glycemic modeling based on the combination of non-invasive biometric data, pump data and a few manual blood glucose values was, in 7 patients, able to considerably increase time in range and decrease time in hypoglycemia compared to results obtained by an experienced team with CGM use.