Post 2 - How In-home Data Collection Completes the View of a Sepsis Episode
Heads up: No clinical advice is being offered.
This is post 2 of a series specifically targeted to in-home sepsis detection. I have a personal stake in making sure I have such a capability – my wife has survived two sepsis episodes and is at high risk of contracting sepsis in the future. Post 1 is found here.
“Sepsis is the body’s extreme response to an infection. It is life-threatening and without timely treatment can rapidly cause tissue damage, organ failure, and death. Sepsis is a medical emergency. Time matters.” - CDC
In post 1 we showed how vital signs captured in-home can support patient and caregiver efforts to measure the signs of risk. Vital signs are easily measured and there are many thousands of measuring devices available at reasonable price points. Christy Health Inc. is where you can capture, manage, and analyze your vital sign data.
In this post we follow-up with data that is not easily captured, managed or analyzed but can be even more insightful than the vitals data. We look at blood lab results. For those with chronic conditions, labs are a recurring fact of life, and when in hospital, labs may be taken several times a day, as will be seen below. When in-home, labs are taken periodically, as ordered by doctors, by visits to a lab, home collection and shipment to a lab, or in a doctor’s office for instance. In our case, it’s rare to go more than 30 days between blood samples. It’s not uncommon to have blood drawn 5 or 6 times a month.
Once again, this data is not well managed by anyone other than ourselves. Most healthcare providers have a website for patients to look up results that organization produced. However, Christine is currently using four lab companies depending on who ordered what test. This means, to see the ‘whole’ picture one must aggregated the lab results of all the lab providers. Over the last 4 years Christine has blood lab results from eleven different companies (including four hospitals).
In post 1 we didn’t group the data to account for where the data was captured: in-hospital or not in hospital (in-home). We simply looked at all the data, over time, to get a picture of the whole sepsis episode. Now we want to introduce the idea of a general proxy for the overall health condition of a person; that is, when the person is in or, not in, the hospital. This is useful because it gives us the ability to define two states in time by means of a simple rule: in-home → more healthy, in-hospital → less healthy. We have the timestamps for each reading so grouping in one or the other of these states without overlap is simple. With data grouped in this way we will be able to compare the measures of the two states, make and test hypotheses. We’ll do a little informal hypothesis testing in this post and we’ll get more formal in later posts.
quantile lines and the black dashed line is the patients median reading. The thick gray horizontal lines mark the labs high and low range markers. As can be seen, Christine’s long term, more healthy, 90% and 10% quantiles are well within the lab ranges for the general populations ‘normal’.
What we can say with high confidence is the AST measure on the 12th is not an error and didn’t jump on a single day. But this begs the question, how did AST evolve from 16U/L on the 7th to 58.5U/L on the 12th? We’ll come back to this question later in this post.
Another question goes to our investigation and analysis of the first possible date of sepsis contraction. If AST has any explanatory power to help answer this question it will be useful. In addition, if other lab results give corroborating results, we will be able to strengthen the statistical inferences used to make our conclusions. We will come back to this question in a later post.
The Pictures of a Sepsis Episode
The dates of interest are:
It’s easy to see the normal range of AST for Christine (10% quantile of 11U/L to 90% quantile of 26U/L) is well within the lab produce range (5U/L to 34U/L) for Christine’s more healthy state. On the other hand, Christine’s AST range is well outside the lab produced range during the less healthy state, from February 12th to February 21st.
Following up on a question posed above, how did AST evolve from 16U/L on the 7th to 58.5U/L on the 12th? There are many statistical technics that can be applied to this question. We apply the straightforward method of backward shifting a moving average to estimate one potential path. The basic steps of this method are as follows:
This helps us in several ways. First, being able to show ED and ICU staff both the results and timings of the labs and vitals data is immediately helpful for building specific patient context. Second, it helps us estimate the evolution of infrequently tested data points when working with fast changing health episodes like sepsis. Note, that once in-hospital lab tests were performed daily at a minimum until the patient was showing improvement and discharged. We can go back to the total timeseries (left most graph) to see that by early March, Christine was back in to her normal range. If you’re wondering about that bump up in later March, yes, we know what it is but don’t want to get side tracked from this set of posts about sepsis. Stories within stories and the data shows it.
Lastly, we have an additional way of determining the first possible sepsis contraction date using blood lab results, in addition to the method using vitals readings. These methods will be combined in a later post examining the estimation of the sepsis contraction date.
We augment the pictures for labs with more history because of the sparseness of the data relative to the vitals data. The new picture sets were explained and used above. We have sufficient data to produce 55 timeseries for unique lab tests over a four-year horizon aggregated from seven lab companies. Which, is too much for this post. We limit ourselves in two ways: the long-term time window is from July 1st, 2017 to April 19th, 2018, and to 11 lab test results. The test results are limited to handpicked, interesting cases coming from the standard blood tests, CBC, Basic Metabolic Panel, and Differential. In the interest of brevity, AST was already shown, and we don’t show RBC, hematocrit, MCH, MVP, glucose, potassium, chloride, carbon dioxide, anion gap, urea nitrogen, polys, eosinophils, and basophils.
For those not familiar with the test names, abrivations, or meaning, this is a good reference site https://medlineplus.gov/ , and use the search box. We let the pictures do the talking.
Red blood cell count and hematocrit follow the same pattern with similar range profiles as hemoglobin and are not shown.
MCH follows the same pattern with similar range profile as MCV and is not shown.
There are a great many observations that can be made from this data, and we’ll make none of them here.
Along with post 1, post 2 demonstrates the possibility of viewing episodic health events by combining in-home and in-hospital data. We’ve shown a few lab results collected both in-home and in-hospital. The lab work shown above was performed by four lab companies and combined into a single timeseries for each test.
The next post will combine vitals and lab data to introduce a Sepsis Risk Indication module and to show an analytical method for retrospectively estimating a date of sepsis contraction. Being able to estimate the infection date of sepsis episodes will be an important contribution to the current state of understanding sepsis.
Subsequent posts will examine how patient introductions to ED staff can be greatly accelerated by use of the graphical history, the importance of individualizing the patients’ normal readings relative to the populations ‘normal’.
Sepsis is a complex process and we use these posts to help build awareness, support those in similar positions as my wife and I, and to offer our services as we are able.
 MedlinePlus https://medlineplus.gov/labtests/asttest.html
Mitch is Christy's caregiver and husband. I've had to learn fast because she really keeps me on my toes. I've found that organizing her data and health information, while challenging, is the only way for me to begin to understand how complicated a life she lives.