Insulin dosage calculation

since about 2 month i’m using an insulin dosage calculator. And I must confess I’m hooked. it took some days for the implemented AI to learn my habits but after that it correctly tells me the optimal dosage. Its an Android App and you need a CGM (e.g. Libre).

Over years I was constantly fighting high and lows and even after years of learning how to estimate the right dosage I still do it wrong. A pump did not help either. I found to main reasons for it: 1) my lifestyle: I’m constantly on the road where each day is different 2) simply miscalculations: its just incredible complicated to calculate the dosage considering amount of carbs, insulin already in your blood stream, sports, sleep and more. My doctor already told me the frightening consequences.

Long story short. I have the app on my smartphone. It connects to my CGM so I can use it as an glucose monitor. To use it I do the following steps

  1. I measure my glucose by scanning the CGM
  2. I add the amount of carbs I eat
  3. I hit a button and the app tells me the amount of bolus insulin I should take
  4. I confirm it and take the insulin
  5. done, that’s it, takes less than 30 seconds

Before you ask - it works surprisingly well. These frequent up and downs are gone. Life became a lot better.

So what is the magic? The tool has been developed by diafyt MedTech, a Germany based startup. It took them 4 years R&D with money from the EU and private funding. FDA just started and the certified App will be on the Appstore in about 3-6 Months for a closed group of users.

Let me ask - what are your thoughts about insulin dosage calculators? Would you use it? What are your concerns?

Thanks for your replies. -Thomas

Every insulin pump calculates the dosage of insulin exactly how you describe. Nothing new. They just developed one as a stand alone app without a pump. I don’t think you need a CGM for some of them.

I would not use it without knowing how it operates.

You mention “… what is the magic …” but then fail to provide any answer or insight.

I hate black boxes.

I also do not know what “FDA just started…” means.

I would also like to know what is your connection. You mention this app is not yet available although you are using it. It would seem that you are a promoter of sorts. Without knowing your connection, it is difficult to have any degree of trust in anything that is being posted.

But that is just me.


My thoughts are that they are all problematic because carb counting and assuming you can dose perfectly according to ratios are flawed concepts. Can it account for my stress level and how much additional insulin I need because of stress? How much of the dose is going to be destroyed by the immune system upon injection and the absorption rate at the specific injection site? How fast the carbs will digest and how many will digest now and later, changing the amount and timing of the dose? These are only a few examples of how complicated dosing is and why a bolus calculator that can’t account for all of these unpredictable and unknown biological processes will still lead to highs and lows.


I eat exactly the same measured amount of food of the same type every single meal every single day at exactly the same time and have for many years now, and I also have exactly the same level of activity every single day, and despite that, my blood glucose levels vary wildly. A major factor increasing blood glucose for me is the weather, and a sudden onset of cold weather can increase the values enormously, so for example, I am now taking a total daily insulin dose 30% higher than two days ago because of a recent cold snap, and even with that dosage increase the blood sugar levels are still way higher than a few days ago.

I suspect many others may have this same problem with spontaneous shifts in the body’s insulin requirements depending various hormonal changes we can neither measure, control, nor anticipate, but they don’t see it that clearly because they have not eliminated every single other variable that could be causing changes as I have.


I also have a very consistent diet and lifestyle, yet still sometimes have numbers that are outside of my desired range. I recently upgraded to the 670G, which uses a closed loop system (in case you don’t know what that means, it basically means it automatically & continuously adjusts certain parts of your insulin based on your CGM). And this by no means eliminates the need to count carbs or occasional highs & lows, but it does help adjust for the small differences between each individual day, since as we have said, there are factors that are virtually impossible to include when calculating your insulin. And it is AI, so it will take a few months before everything is tweaked (both settings & giving the pump time to learn how to optimize for you). I still usually have spikes after breakfast, but they peak at much lower levels, and most of my day is much more level. I think closed loop systems (whether the 670G like me or another one) are something all diabetics should try. No diabetics’ numbers will ever be as good as everyone else’s, but this is probably one of the biggest and most significant new technologies available to us.

The problem in the US is the patents. I use the Omnipod pump, manufactured by Insulet, which has a built-in calculator but the model the calculator uses of insulin adsorption is bogus. This is because Medtronic has a patent on using the obvious improvements - insulin adsorption is well understood - and Medtronic was busy suing Insulet anyway last time I heard anything.

If the app you have been testing does actually have a good model of insulin adsorption or if the designers have managed to find an end-run round the Medtronic patents by not using a model but, rather, using the CGM reports to effectively reverse engineer one on the fly then it is very definitely an improvement over the lousy insulin pump algorithms.

BTW: reverse engineering insulin adsorption from CGM glucose levels is a novel and patentable idea that I just published. It is not and cannot be patented by anyone now. If diafyt MedTech are trying to patent it they had better have already submitted something.

Personally I would recommend open source solutions; while technically they cannot break the Medtronic patent stranglehold they can use the provisions for individual experimentation to publish stuff that would get a private corporation sued. I’m not greatly enthusiastic about start up vaporware.

IRC there are a number out there but I have yet to use one.

John Bowler

what the insulin pump does is much simpler. Its more like a 1980s calculator. diafyt uses a mathematical model of the diabetic metabolism and machine learning to adapt to your personal habits. This makes it possible to replicate the body functions and restore the natural control of glucose.

would you ask the same question (how it operates) if it would come from Dexcom? But since we are not Dexcom I can answer your question about your black box concern. diafyt consists of to parts 1) an algorithm that simulates the human diabetic metabolic system and 2) a machine learning for parameter tuning. Both sits on Amazon cloud. The App calls the service and displays the recommended dosage.

Talking about your FDA question: you are correct. without FDA certification nobody can use it- in the US. But since diafyt is in Germany things are a bit different. We are doing CE for medical devices, FDA and CFDA. But please be patient. This takes a while.

My connection: I’m a cofounder of diafyt, so yes I’m a bit biased. But trust me a bit. I’m a T1D myself, I’m super skeptical and a scientific tester. And if you need data … in 4 months I could lower my A1C from above 8 to below 7.

good for now? otherwise just go ahead. I’m happy to address all your concerns.


you are absolutely right - “carb counting and assuming you can dose perfectly according to ratios are flawed concepts”
So whats different? diafyt understands your diabetic body as a system, not as a set of individual body functions. T address that we build a simulation of the metabolic system and run your actual data (glucose level, time of day, calories burned, …) through it. On top we added a machine learning to tune parameters and adapt to your lifestyle and individual condition e.g. insulin resistance.

hmm, I’m afraid I cannot make any therapy recommendations so I guess talking to your doctor about your problem might be the best way.

May I ask what is your A1C, are you a T1D?, can you send a graph where I can see the out of range levels and times?

Regards diafyt, it would be an interesting test case.

long story short. We spend about one year to run existing algorithms through MATLAB to learn why these algorithms are flawed, eventually deciding to start from scratch not relying on existing patents or reverse engineer.

Also we already filed a patent and there will be more to come.

We will provide an open API where others can build there own solutions and integrate our insulin calculation.

So in this context diafyt software is exactly that what you need to build a closed loop system. However we wanted to build a super simple tool that does the same job however costing a fraction and is good for people not using a pump.

diafyt needs less than 10 days to learn and is more stable even in unexpected changes during the day. There is nothing to tune. Its all automatic, just one button to press.

For the first week I was super skeptical trusting the calculation so I did overwrite the recommendation with my own best guess learning that the recommended dosages would have been better than my prediction. So after a while I started to follow the advise in almost all cases. And voila glucose levels are better, less high/lows and less to worry about.

diafyt is a Germany based startup with a team of software engineers, T1D, PhD in Mathematics and supporting endocrinologists. I’m one of the founders.

btw. we presented our technology at the Diabetes Technology Conference in Washington and the Digital Diabetes Conference in San Francisco.

I have no idea why such a statement. Being the first statement in your response, it colors your post. Unfortunate. One can only assume you have had very bad interactions of some sort with Dexcom. Why you would feel the need to bring that up here and make that the first sentence in your reply to me is beyond my comprehension.

In my opinion this is “Black Box” in that we would never know how or why the resulting calculation is what it is. To that degree, you have not answered the question but merely reinforced that this is indeed a “Black Box”.

That simply says “where” the calculations take place. The calculation does not take place on the app on your smartphone but rather an internet connection is required and larger servers elsewhere calculate the result and send the resulting answer back down the internet connection to your smartphone. Mildly interesting. Currently we are not dependent on wifi connections in order to bolus. I doubt we would want to become dependent on wifi. Currently we use wifi when it is available and if not - well - no problem.

That still does not answer the question. “… doing CE …” and “FDA just started …” simply sounds like vague implications not actually back up by what the statements imply. Some companies will provide concrete answers about their current interaction with the FDA.

You had an opportunity to state your connection in your original post. Not doing that upfront is shady.

Nah - I am all set.

Note that I am a big fan of consumer choice and options. My only suggestion to you would be to rework your original presentation.

I still don’t see how it can account for absorption variances at different injection sites, the immune system response to an insulin injection, or how quickly my body will decide to digest lunch today and how to take and time the insulin for that - these are unknowable things until they happen. I think the bigger barrier is getting people to log data though which is a huge pain. On a pump everything is logged without any extra steps. If I’m on MDI I don’t want to have to pull out my phone every time I take an injection and have to enter information. I sometimes take up to 12 injections per day and multiple injections with a meal. I’m often with other people and quickly/discreetly bolus with my pen under the table or while out somewhere. Unless the app can communicate with a Bluetooth pen and bluetooth CGM I would have no interest in it.

diafyt has one mission - reliably calculate insulin dosages, no more no less. Its for people who struggle with frequent high and lows and high A1Cs. Its simple to use and you do not have to have a pump.

So maybe its not for you. However “I sometimes take up to 12 injections per day and multiple injections with a meal.” would tell me another story. Your frequent bolus could already be an indication that your calculation was wrong and you need to add a correction bolus. From my experience with diafyt these extra injections go away. . its just a thought

im not selling you anything (yet). diafyt is in closed Beta and FDA just takes time. Being a tester, I only wanted to share my excitement about how it helped me managing my life with diabetes. My A1C is better, I have less high/lows and it actually takes less time to manage it.

So if you are already happy with your diabetes management this might not be for you. If you have challenges just keep following us and do sign up for the next closed test group as soon as it comes available.

Nope, it’s not miscalculations. It’s using my pen like a pump and having to deal with spikes caused by stress, protein, and food that digests later but can’t be accounted for or bolused for in advance without going hypo. This is where AP systems do make sense because they will give more insulin in these situations, but on MDI this means more injections, not less.

Eric to me it sounds like mis-calculation if you are doing multiple injections during meal time.
Meal time isn’t hours long so if you start your meal and know your expected carb intake you should be able to bolus (take 1 shot)

I for one wouldn’t mind trying this diafyt because after 50 years, 100 of thousands of injections, many pumps, all types (brands) of insulin, and many CGMs, and working with specialists Europe and here in the States, I still have highs (200-350) and lows <50 if not daily at least several times(4 or more) a week, doing the same thing over and over and A1c never below 7.

If we were MDI, we would absolutely be doing multiple injections. Food takes a considerable amount of time to digest depending on the type of food (carb/fat/protein). We use a Tandem t:slim X2 pump and make frequent use of the extended bolus feature. On MDI without extended bolus, we would certainly be using multiple injections for a single meal. And that does not even take into account deciding to have another serving of something. Or following up with desert.