Have data, will recover
Chemotherapy for blood cancers shares many of the joys of other chemos, with a little twist. Because it's the blood producing mechanism that has gone haywire, chemotherapy agents leave your body unable to produce key blood cells such as your red blood cells (what carries oxygen), platelets (what helps you clot when you bleed), white blood cells (collectively involved in fighting infection), and the important subset of cells called neutrophils which are the main cells associated with your immune system. Your immune system is knocked to measurably zero. In this state, you are so vulnerable to infection that at the UCLA blood cancer ward, you are not allowed to take showers. There could be aerosolized pathogens in the shower. Instead, the medical staff draws a bath, pours in disinfectant, waits, and only then you can bathe. Normal toothbrushes are contraband; your platelet levels are 1/10th of normal so toothbrushes are a bleeding risk and normal oral bacteria can lead to blood infections (happened to me twice). So yes, the point at which regular hygiene is a health risk, you can imagine you'll be well isolated in a hospital for a while.
There I was, in hospital isolation for a month at a time for each of the 4 chemo rounds. No children under twelve, no leaving the ward, no fresh fruits or vegetables (there could be bacteria on the surface), no walking outside of your room without a mask. They take a blood draw every morning to track your counts to determine when you need red blood and platelet transfusions and monitor the progress of your white/neutrophil counts until you have enough to be discharged. Your temperature is taken every 4 hours to check for fever. With no immune system, the first and last sign of danger is a fever. As soon as your temperature crosses 100.4 F, the parade starts with a blood culture, chest x-ray, and an antibiotics regimen that must start within 2 hours and will last for the remainder of your hospital stay.
It's the ultimate wait for the shoe to drop... Or is it? Blood counts, numbers on a chart, data. I'm an engineer who was now stuck in the hospital with not much to do. You bet that Excel sheet was going to come out.
If you've ever tried to get information out of a doctor, you know how it goes: "everybody is different," "it's hard to say exactly" and then they will cite general rules of thumb. For the first chemo cycle, there was no other information so I relied on their mantra: "platelets are the last to come back" and "your neutrophils should be coming back soon." Both turned out to be wrong: my platelets came roaring back first and my recovery took 4 weeks instead of what we discovered later to be their unspoken expectation, 2 weeks.
But with each subsequent round, I have more data. The human brain is great with pattern recognition. And I'm also starting to learn my own signals: an elevated blood pressure is usually correlated with a fever. If I start to feel cold outside of 2 pm, when the HVAC system would go into overdrive, there was a good chance I would be fevering in the next 1-2 hours. The attending fellow became a believer when I told him that based on historical data, I expect to spike a fever sometime in the next 48 to 72 hours. "We'll see, you're doing great" he incredulously said. 48 hours later? Fever. The tables turned. Now the doctors were asking ME for MY predictions of when my counts would rise and in what time range I would be discharged from the hospital. I flagged major events in the spreadsheet and could pull up how long I was on which drugs faster than the nurses.
OSIsoft makes data software for industrial operators. We hear about the “asset whisperers” all the time. The guys & gals that know their assets and processes like the back of their hand; they hear a whir that's a little off or notice a fluctuating reading and can predict an impending equipment failure. I have always been impressed and now have an even greater level of understanding for just how far big data and all these other initiatives must go before catching up to the human brain.
Because the challenge is to bridge that gap between the aggregate and the individual. In that copious quarantined time (before the rest of the world was also quarantined), I combed through scientific papers. But it was hard to apply the medical data. How similar were these patients to me? How old? Did they have the same genetic mutation that I had? What chemo agents were used? Was their bone marrow similarly slow to recover?
Industrial operators face similar challenges trying to characterize and predict failures of their own assets. Key data is scattered across systems, and the meta data isn't always there to make an apples to apples comparison: have the other units been running in similar weather conditions? Workload? Is the longitudinal data available to overlay the baseline degradation expected over time? To make representative models, big data is often not enough; it has to be HUGE data. And well curated huge data. Huge enough that there's enough data left in the training set after it's been segmented. General statistics are absolutely better than nothing, but I also experienced the gap between those distributions and my own ability to recognize patterns. As everyone builds up their data sets and the buzz of advanced algorithms fills trade magazines, it is easy to overlook the nearly prescient insights of the asset whisperers.
One day, we'll have the datasets to be able to slice and dice on the relevant variables that will allow us to create reliable models. We'll get there. Granted, HIPAA regulations will make that tougher for medical information, in addition to a lot of the meta-data being stored in typed note format that is difficult to extract. I would LOVE to see more data sharing in general as we can especially appreciate in these COVID times as we try to understand the risk factors that are applicable to ourselves; will we be part of the lucky asymptomatic group or do we need to brace for getting hit hard? Heavy process and manufacturing companies are certainly further along, and equipment vendors are making inroads. The rise of IoT devices certainly lowers the cost of collecting the data although we've still got the consolidation and normalization to contend with for now. It's going to take a large quantity of data, good tagging / context, and a means to work with both validated (e.g., from a hospital or control system) and informal (e.g., from individuals) data. Big data and AI will get there, no doubt.
In the meanwhile, I'm continuing to bet on the engineer with some rag-tag tools and a vested interest in getting the prediction right. That seems to be what my doctors did.