The Crucial and Overlooked Step in mHealth Trials
What is necessary for an mHealth trial to be conducted? I saw an article recently that talked about devices, recruitment, outsourcing application development, but failed to mention anything about data analysis.
I have worked on many mHealth trials, and one of the crucial things sponsors often miss is data science and having a comprehensive data analysis plan for mHealth data. I think this happens because pharma organizations are used to handling data generated from clinical trials (and even some nonclinical related sources), and so they think that the techniques and processes that were used previously can be applied to an mHealth trial. This is definitely not the case at all, and if you ever plan to undertake or plan any future mHealth trial, data science and analytics should be your first step.
Data is not all the same, and there is no standardization
First and foremost, the data coming in from mHealth trials is all different. I have seen provisioned devices in combination with ePRO, provisioned devices alone and even custom apps that have ePRO plus ill-designed, half-baked tests that use sensors on the phone. In none of these cases were any tests done as to whether the data coming in would be effective for measuring the outcome, nor was any planning done in determining how to analyze the data. They just assumed they could, and that the data would be useful.
If you are doing a study that includes non-PRO data (i.e. a device or using phone sensors), run tests to figure out the data format, how the data will arrive to you, and at what frequency. I can’t tell you the number of times I have heard of cases where organizations assume a device vendor will just give them data; in a lot of cases, the device vendors are small and do not have their operation set up for data extraction requests.
Secondly, there is no data standardization. Accelerometer or heart rate data coming from one device will definitely not be at the same format or interval frequency as another. If the sampling rate isn’t good enough, it will compromise your trial because it means the data you are collecting will not be sufficient enough to explore your endpoints. It’s vital that you thoroughly vet devices and make sure you get sample data outputs to confirm you can use the data.
Focus on a patient engagement strategy, and account for mistakes
You should account for dirty data. Count on patients to make mistakes. Sometimes they do the tests incorrectly, sometimes they don’t do them at all. The key to mitigation is having reports in place (or even a dashboard) that can alert you to compliance issues and non-adherence. Having open communication and transparency with the sites is key to mitigating a lot of these issues, as the sites can communicate with patients to adhere to the instructions.
Treat data science as a first class citizen
Most importantly, treat data science as the number one priority in your plan, not number five. Assuming all the above steps are in place, and you have a steady stream of good data coming in, count on using modern tools to analyze the data. We typically use R and Python in combination with Docker on compute clusters to analyze bigger forms of data in a scalable approach.
Analyzing mHealth data is a relatively new science, so count on a lot of experimentation to garner insights into the data that ultimately leads to proving your endpoints. We read a lot of research papers in this process, and we are constantly looking for new ways to analyze and get information from data.