|
Speaker
|
Transcript
|
| bnossaman@ochsner.org |
Okay. |
| |
and |
| Nicholas Shelton |
You can begin when you're ready what I will do is just ask you to just watch the time, so if you can keep it around 20 minutes that's fine. |
| |
And, and if you make a mistake it's not not a big deal just just keep on continuing with the presentation, they can always make any edits if necessary later on. |
| bnossaman@ochsner.org |
Thank you. |
| Nicholas Shelton |
Okay, when you're ready. |
| bnossaman@ochsner.org |
You can start at. |
| |
905 right. |
| |
Thank you this morning for allowing me to present this interesting conundrum that we had in our intensive care units at Ochsner Health in New Orleans at the beginning of the COVID epidemic. |
| |
We did not have, nor did the nation, maybe even the world, have a design protocol to address this life-threatening pandemic, other than supportive care. So I began designing |
| |
an informed decision analysis which the results will be presented this morning, to see if we can tackle this ball that is rolling towards us. |
| |
At the beginning of March 8, 2020, |
| |
we saw the sudden, tremendous spike of COVID-19 patients. Along with that was a tremendous increase in the number of patients requiring mechanical ventilation... |
| |
pulmonary ventilation of the lungs, okay. |
| |
I'm fortunate that I am an intensivist and was involved in the care of these patients, although we had at times upward to seven units dedicated to care of this specialized patient population. |
| |
Preliminary calculations said that |
| |
I would probably need about at least a 200 patient database, which to make some kind of formulations on how we should approach this patient population. |
| |
As we see in the first static slide, this was the mortality distribution in this patient population. |
| |
And we see a dramatic increase, beginning at age 50 upwards to almost 100 years of age, of the high hospital mortality rate in this patient population. Of course, these data can be summated either with a median at the entire quarterile ranges or a mean of the standard deviation. |
| |
But in the end we wanted to do some kind of prediction on what type of patient that would arrive at our doorstep, and more importantly, |
| |
how to manage the care of these patients. So even though that we have a prediction interval calculated that the next patient would be within a 95% confidence interval 45 to 99. |
| |
Although this is important information, |
| |
it's not what where our primary intrest is in. So in March 2020, we begin to harvest up to 500 data points in each of our initial 211 COVID-19 patients who received invasive positive |
| |
pressure ventilation within the Ochsner Health Center in New Orleans, Louisiana, in about four weeks, which is basically this range right in here of patients |
| |
we believe we calculated a sufficient and obtained a sufficient sample size to allow this development of an individualized institutional protocol |
| |
or an informed decision analysis that, at that time, current supportive therapy included maybe the use of hydroxycloroquine, the use of ??? or an anti infective, |
| |
proning patients, i.e., turning them face down, the use of medications to support the systemic blood pressure, the use of hemodialysis for renal failure and the use of anticoagulants, i.e. heparin to prevent the thrombogenetic component that occurs when the body undergoes major assault. |
| |
However, this data extraction required about five months of careful extraction and so we used this for later stages of the pandemic. |
| |
And we now use this database to help us pair future therapies, thank goodness for monoclonals, which we had at this early stage of patient care. So our major outcome of interest is basically hospital mortality, |
| |
which regretfully troubling, was 65%, which corresponds to the same range that was reported in different medical centers in China, |
| |
in France, Canada. Of course, we have the confidence intervals, for those of us that are should present this information, it can range from 58% to 71%. |
| |
Regretfully, in our patient care, although we had a very high mortality rate and unacceptable hospital mortality rate, those patients who did survive to hospital discharge, we did not have a good track record in that area either. |
| |
Only one in 10 patients, really less than one in 10 patients, went back home to restart their day-to-day activities. The vast majority of this patients required some degree of assisted |
| |
help at home or placement in skilled nursing facilities etc., or long-term acute care facilities, although the numbers are small, because regretfully, we had a high percentage of mortality. |
| |
So we then started to examine the etiology of this COVID by organ system. |
| |
Yes, the majority of patients came to our hospital with respiratory complaints, but a number of these patients had other issues as their primary complaint, we call it the chief complaint in medicine. And we see those listed here, and if we activate |
| |
hospital mortality, we kind of get a feel for if you came in with an infectious etiology get a feel using your eye, is this high mortality rate? |
| |
We had two patients with abdominal complaints, which is again 100% mortality, but a small number, nevertheless we kind of get an idea of what we're are dealing with. |
| |
We then looked at admissions symptoms, patients and their complaints. And again we have the big six |
| |
admitting with fever complaining of fever; generalized weakness or fatigue; diarrhea; cough; shortness of breath or anosmia, which is that lack of smell and taste, which was frankly |
| |
a symptom that was reported late, certainly not within the first four weeks, and so we kind of discount this. And again if we just look at the mortality rates, |
| |
looking at these symptoms, surprising to me that we had a high...higher percentage of incidence of mortality in patients without fever |
| |
or without a complaint of cough or without shortness of breath, as compared to the other patients that did present from a percentage basis. We're using our eyes, just to kind of gauge the the volume of regretful mortality in these patients with this type of |
| |
complaints. |
| |
We then looked at ...a busy busy slide, I apologize for the busyness of this slide...but we looked at |
| |
the vital signs assessments, these six parameters that we use, and see if we can detail any information when these patients present |
| |
in our emergency rooms. And we're getting this initial set of vital signs to see if we can use this as a predictor. And again just turning on the mortality rates, we see that |
| |
pretty much everything follows the shape of the histograms that we see here okay. |
| |
We were surprised that a lot of these fresh patients didn't have more of a lowering of the diastolic blood pressure, because, generally within systemic inflammatory response syndrome, |
| |
or SIRS, the diastolic pressure is usually mounting a response in the 40s and 50s, certainly not in these high values here. So this suggests, to me at least, that these patients are suffering more from a |
| |
sympathetic overstimulation syndrome described by Dunser and his colleagues in Germany back to in the '90s. |
| |
SOS, to paraphrase, is associated with a very high mortality. This may have a key role to play. |
| |
And then, of course, a pulse oximeter, which normally in most patients, under baseline conditions, is right around 100, little less than 100. |
| |
And we see that the vast majority of patients came in with very low values, which surprisingly, there were survivors in these low groups, so this has a role to play, but not necessarily a definitive role. |
| |
We then looked at comorbidities and the distributions of those comorbidities. And again, essential hypertension, diabetes, coronary artery disease, heart failure, tobacco, |
| |
history of COPD (chronic obstructive pulmonary disease), reactive reversible airway disease, kidney disease, |
| |
and leading all the way down to organ transplantation and malignancy. And if we turn again to those patients who regretfully died during hospitalization for COVID |
| |
in this cohort, we see that organ transplantation, if you had that, you regretfully experienced 100% mortality. But to some of our surprise, |
| |
that we're starting to kind of get a feel where these patients with essential hypertension, it looks like |
| |
that the percentage mortality is about the same, certainly it is for diabetes, congestive heart failure, maybe less, |
| |
certainly with coronary artery disease, a higher percentage getting a feel that they would have issues with coronary disease and COVID. And tobacco abuse, a high percentage, which maybe will have a role to play, a little bit later on. |
| |
We then conducted a nominal logistic fit for a hospital mortality, because again |
| |
you have to look at all of the interactions that all these comorbidities play against hospital mortality and again, this is a lot of risk factors that may not be meeting the strict |
| |
statistical criteria for modeling with nominal logistic fit, and I acknowledge that limitation. But we see that tobacco abuse is very high, with a low probability due to chance of 97 in 10,000, and as well as renal disease, etc. |
| |
We had to eliminate the organ transplantation, because the counts weren't enough to allow the model to regress. |
| |
We then generat...more interested in doing generalized linear model fits with standardized risk differences, so a fun way to look at |
| |
outcomes data, looking at the patient presentations with their symptoms, and none of these items were quote statistically significant, and risk of mortality wasn't there because all the confidence intervals cross zero. |
| |
We then looked at one of the key features for...that JMP provides is of partitioning, recursive partitioning. And |
| |
doing an internal validation, here are all the 211 patients looking at your admission laboratory parameters, the blood work, if you will, and we split the data and, surprisingly what had the most |
| |
associating with hospital mortality being represented in the in the individualized patient points. Underneath the lines here was a protime or a measure of coagulation, and if it was elevated, you had a higher mortality rate, basically of about |
| |
85%, whereas if you had a protime less than 13.4 seconds, about a 60% mortality rate and the log worth value being over 3. |
| |
So from a statistical standpoint, the pattern is is not really due to a chance; it's possible, but not |
| |
not really. Okay, and then if we split it one more time, we get this procalcitonin, which is a marker in infectious disease literature is |
| |
a sign of bacterial infection not viral, but bacterial infection. And we see some differences being expressed here. Is this of benefit? |
| |
In this ???gram, well, the receiver operator characteristic curves are not as robust as we would like, so maybe this is more due to chance and less due to patient care, but certainly worth exploring. |
| |
Here are hospital length of stay graphs in our patient population, and we see that the length of stay rises as high of about 20, |
| |
just under 20 days and it's all the way up to about 70 days. And regretfully, we had some patients that didn't stay long in the hospital, and if we plot out that information, |
| |
we see that a lot of these patients died very early on. It's regretful. All these patients, like I said, would go into the intensive care unit initially, so we have that information here, being expressed as mortality rates, determinates of percentages, etc. |
| |
So then, it's our first overlay plot, which I would like to show the audience, in which... |
| |
It's a little busy slide at first, all 211 patients between March 8 and April 5, 2020. Look at the patterns and the degree of randomness and, of course, regretfully, if you didn't make it out of the hospital and into the...into the hospital ward, |
