Home / Resources / Articles / How Efficient Are Voice-Operated Apps for Insulin Pumps?

How Efficient Are Voice-Operated Apps for Insulin Pumps?

Feb 10, 2018

Evaluating if bolus calculators for insulin pumps can be accurately calibrated by speech recognition.

Insulin pumps have become some of the most efficient devices for people with type 1 diabetes. Insulin is provided from these pumps through continuous subcutaneous insulin infusions (CSII) and bolus administrations in response to daily meal loads. It has been shown that postprandial hyperglycemia leads to cardiovascular disease causing myocardial infarctions and death. It is important for some patients with T1D to accurately calculate how much bolus insulin is needed to compensate for each meal they eat. This takes into consideration fasting blood glucose concentrations (BGC), targeted postprandial BGC, insulin-to-carbohydrate ratio (ICR), and the insulin correction factor. It has been reported that most people prefer calculating programs that are computerized or an app on their smartphones to avoid calculation errors. Recently, the VoiceDiab system helps these patients with insulin bolus calculations by estimating the amount of carbs, protein, and fat in each meal.


The VoiceDiab is an Android app that works through voice recognition. It calculates needed insulin doses to compensate for meals based on the voice description of the content of these meals. The app is made up of three different servers: 1) automatic speech recognition (ASR) that converts verbal narrative of meals into a text format, 2) analyzes the text format to stratify meals into protein, fat and carbohydrate content, 3) lastly, it calculates compensatory insulin doses for the meal based on a specific algorithm. 887 basic foods are programmed into the app with information about their caloric value and amounts of carbs, protein and fats in 100g of their mass. The language database can decrypt slang language and can be customized to work in Polish and other dialects. When a patient speaks into the VoiceDiab system, it typically takes 15 secs to complete a list of a 3-food meal. The insulin dose calculator then calculates the needed insulin dose in 2-3 seconds to the smartphone. The app stores the patient’s ICRs, type of insulin device, and the maximal number of insulin units that can trigger an insulin overdosing warning message to ensure patient safety.

The purpose of this randomized 2×2 crossover study was to evaluate the efficacy of the VoiceDiab system by testing postprandial BGC, different variabilities of BGC, and the time of BGC levels in and out of a targeted range in patients with T1D who are being treated with an insulin pump. Patients included in this study also had HbA1c levels <10% and knew how to operate smartphone applications. Forty-four patients were randomized into two treatment groups and observed over two separate four-day time periods. Treatment A group used the VoiceDiab system for insulin bolus calculations, while Treatment B group performed manual calculations in the first four-day period.  After a 14-day wash-out period, the second period began integrating the crossover design of each treatment group. Patients were blinded to their glucose levels. Insulin dose (ID) was calculated using the formulation ID= ICR x CE (carbohydrate equivalent) + ICR x PFE (fat-protein equivalent). When analyzing food, 1 CE can be given to a food item if it had 10g of carbs. 1 PFE is also equivalent to 100 kcals of energy, anything below this amount would equal 0 PFEs. Insulin administration was given in a dual-wave bolus when protein and fat content surpassed 100 kcals. BGC was monitored using Platinum G4, a continuous glucose monitoring system. T-tests were used to analyze results.

Results showed that on average, the VoiceDiab system was used 6.1 times per day by patients. In Period 1, Treatment A group used the VoiceDiab system and showed a lower percentage of postprandial glycemia (~13.2%) in comparison to Treatment B group who used manual calculations (16%) (P=0.028).  In Period 2, Treatment B group used the VoiceDiab system and showed a lower percentage of postprandial glycemia (~12.9%) in comparison to Treatment A group who used manual calculations (16.9%).  Overall, there was a significant difference in controlled 2-hour postprandial glycemia between VoiceDiab system support (58.6%) and manual calculations (46.6%) (P=0.031). There were no differences in other analyzed parameters.

In conclusion, newly invented software can help reduce insulin dose calculation errors.  This can lead to better diabetes control and an overall better quality of life for patients.  Use of the VoiceDiab resulted in a 12% difference in BGCs in studied patients. Patients were also surveyed after the study and provided positive feedback on VoiceDiab’s usefulness, effectiveness, disadvantages, and advantages. Some disadvantages of the app were the lack of certain food products and manual data entering errors. Patients also reported that their manually calculated insulin doses differed from the VoiceDiab system’s calculations. More than 87% of the patients reported they would use the VoiceDiab system in the future. This study lacked generalizability because it focused on a younger population. Future studies should increase their power by increasing the population size and broadening population characteristics.

Practice Pearls:

  • Accurate calculations of bolus insulin are important to avoiding detrimental effects caused by hyper or hypoglycemia.
  • Many patients prefer computerized programs and phone apps to track their foods and insulin requirements due to accessibility and accuracy.
  • VoiceDiab system provided accurate insulin calculations and better BGC control.


Foltynski, Piotr, et al. “Efficacy of Automatic Bolus Calculator with Automatic Speech Recognition in Patients with Type 1 Diabetes: A Randomized Crossover Trial.” Journal of Diabetes, 8 Jan. 2018, doi:10.1111/1753-0407.12641.

Adrianna Jackson, Doctor of Pharmacy Candidate: Class of 2018; LECOM College of Pharmacy