Lecture Transcripts

Lecture 1 Transcript Welcome to this first lecture in the course on clinical cams stream. In this lecture, I'm going to cover the topics in chapter one where we're looking at a general overview of what clinical chemistry is, including molecular diagnostics and placing it in the field of Laboratory Medicine in general. So first of all, what is chemistry? Clinical chemistry is really the area of chemistry that's concerned are focused on the analysis of bodily fluids. With the goal of making an assessment for diagnostic or therapeutic purposes. A medical or clinical laboratory then is the place where these tests are taking place? January. It's clinical pathology lab where there's going to be other test being carried out as well. But the clinical camps she tests, we'll be looking at. The specimen is obtained from patients in order to aid in the diagnosis, treatment, and prevention of disease. In the textbook, they give the following definition for laboratory medicine. It's a component of laboratory science that's involved in the selection, provision and interpretation of diagnostic testing of individuals. It's not so different from the one on the previous page that I really like how the first deformations I provided made it very patient centric rather than just focusing on the specimen because there really is ultimately about the patient. And the clinical chemistry in the textbook is defined as the largest sub-discipline of Laboratory Medicine. Which is true. One of the other major disciplines would be clinical microbiology, or they're looking at bacteria and their interactions with antibiotics. And it is a multidisciplinary fields for sure, including hematology, immunology, clinical biochemistry, and many others. Laboratory testing as defined in the text. And I'll just point out this is at the start of the chapter. So each chapter shows some key words and definitions. If you are looking for those when you're going through your Textbook review. Anyway, here it shows laboratory testing is a process conducted in a clinical laboratory to rule in or rule out diagnosis. Or maybe to select and monitor disease treatment to provide a prognosis, to screen for disease, or to determine the severity of and monitor a physiological disturbance. So typical tests might be blood tests shown here where your doctor may order the screen just to see that there's nothing abnormal going on even though you aren't complaining of a any symptoms. Or they might be ordering this because you are complaining or something and they want to just see if could be a problem with your kidney or maybe you're low in for high, high level of urea and your pledge or any number of things. So here I just show some routine, some some common routine blood tests that are carried out and we have the complete blood count where we look at red. Cells, white blood cells, hemoglobin, hematocrit, platelets are the basic metabolic panel where we have well, that's just saved BMP, electrolytes, calcium, glucose, sodium, etcetera. Comprehensive metabolic panel where in addition to these elements and electrolytes, ions and gases, you add in some other proteins and enzymes. Lipid panel that lets get our cholesterol. And if they're concerned about your liver, kidney liver panel and similarly with the February. So here I show a typical example of a test that you might receive from your screen. So this is a basic metabolic panel. And here we show, for example, the glucose result is 78 and the units are milligrams per deciliter. And then they show the reference interval of what's considered an acceptable range values. So here we see glucose 78 fits nicely in the range of 65 to 99, no problem. However, if we go down here to sodium, we see this is in bold and there are some marking in column. Flag has been flagged as low. So when 32 millimoles per liter is in fact a little lower than they accept reference interval or standard reference interval for most patients. And again for chlorine, they show the same thing. So that's quite typical that if there is a value that comes out of a result that comes out that falls outside of the reference interval, it will be flagged, which drives the clinician's attention to something that may need to be addressed, whether it's measured, again, monitoring every few months or it fits may be indicative of something more

serious. And so that brings us to why do the tests, why do test? It is not only just to figure out what the value is, but I can't be as I just stated to confirm or a clinical suspicion if after the physical exam, Dr. suspecting that maybe you have some kidney problems going on. They might be able to make a diagnosis based on those test results or could be to exclude that just to make sure they might not think is that it might just be sure that it isn't. And so they'll run the tests. It could be for assisting in treatment selection, for example, if you have the brca gene, My first test for that. If the brca gene and if you do it might then determine which type of treatment is best suited to your type of breast cancer. Or it might also tell you then your prognosis and might be different depending on whether you test for this gene or you don't. And we also test in order to screen for disease in the absence of any clinical signs or symptoms. That's typically was done at our annual physical and for establishing and monitoring this very day severity ever physiological disturbance. So they'll history. So interesting to me that clinical Can she really dates back to ancient Egypt. Ancient Egypt really were already than doctors were noting that the urine of a certain group of patients who are feeling ill will sweet. And it was really the early diagnosis of diabetes before we knew what we know today, they would actually taste it. And in some cases it anyway, I think there were cases where they noted the ants were drawn to this would be like an ant test. But of course today we have other ways of measuring that, that glucose, which would have made it taste sweet in the year. Anyhow, the first clinical labs in the US will not established until 1895 course Kwame. There were some labs earlier in Europe, but it wasn't really a system that was being used across the board yet. In fact, early on that many doctors who felt like the chemistry didn't belong in the field of medicine, that they should just be able to assess the patient through their physical exams and questioning. And eventually it gained more acceptance. And during World War one, women were recruited to work in the labs because the men were off at war and it actually became one of the fields that was really female dominant. At that early deaths were women get a job in medicine and science if they were interested in these topics, subjects woman then otherwise couldn't. Now today of course it's, it's very m equal, much more equal in terms of genders representation. But it's interesting to me that it was very female dominant role, the dominant role field in the early diss any app. As time progresses, new our diagnostic methodology is being introduced and I will discuss that also on the next slide. But just like for example, it was not until the seventies that immunochemical diagnostics were introduced. Allergies and antibodies, as well as even more recently in 1987, molecular diagnostics, for example, the PCR test that's being used for coronavirus testing. So what does define the boundaries of clinical chemistry then? Really, where is the limit? Well, basically, it's, requires advances in technology and in our understanding of what a change in a certain analyte will mean and how sensitively we can measure it. Say in the forties and fifties, there were some graded masses and spectrophotometry, which we're going to cover in the next set of lectures. Next week, electrochemistry and Chromatography, we're also going to cover in the future lectures. And then in the 19 seventies, as I said, I mean a chemical techniques were developed and then approved by the FDA. In eighties, mass spectrometer came on the scene. And that made for very sensitive testing of urine, like drug testing. Automation of the systems enabled high throughput screenings and measurements. So you could look at, instead of a few 100 patients per day, thousands of patients. And miniaturization really enabled the point-of- care testing that is starting to be developed now. And molecular diagnostics where we use nucleic acids, DNA or RNA. Amplification techniques emerge to study infectious diseases or act or soldiers or even like the coronavirus flourishes talking about. Okay, so how's clinical chemistry practiced? What are the functions of the laboratory professional? Is it just to run the test? Now, there's a lot of other aspects to what would be done by the clinical games, including the development and validation and new lab

badge, the fight is in S that are looking at blood transfusions from plasma donors should recover from Cove it and they have to be able to analyze the antibodies that are in there. And now they're working with and of course, the active virus we need to know to and how long we have an active virus, how long we have to stay in isolation. And these are all really important tests that are being done and develops as we speak for. So you can see it's really an important field. And it's also a really continuously evolving field. And I hope that you enjoy learning more about it as we go through this course. Lecture 2 Part 1 Transcript Welcome back to the second lecture in this first week, where we're going to be covering the topics in chapter two, analytical and clinical evaluation methods. Now this chapter really is about the language of statistics that are relevant and necessary for clinical chemistry. And although I know statistics is not often people's favorite subject, it really is critical in terms of being able to talk about clinical chemistry in a meaningful way. Because it's really a lot. All of it has to do with the the reported values that we're going to be observing and how they fit into the population statistics and their meaning in the relation to reference intervals that had been determined for a sample population. So we're going to need to understand the statistics for just that type of discussion, as well as the fact that and the clinical lab, you are going to be comparing assays Constantine, whether it's from one lab to another or new instrument that's come into the lab or even for internal quality control purposes. We need to understand a little bit about that too, just to have a better understanding of clinical chemistry in general. So in the texts they go through this flowchart of how we select passe best suited for our goals. So if you may be looking at a new tests or diagnostic that is being developed in a lab, or perhaps just a selection of one that's available commercially. But you'll be doing this in response to having established the need for one perfect example lately is this need to test for by rule of the sars coffee to virus. And a number of labs had been developing their own testing for that. So you need to define quality goal, how sensitive doesn't need to be mz, and what's the best method to adjust this question is going to be PCR in this case. And, or some other technique. We're going to need to have a good method for verifying the results that we're getting and validating the results for a new test. And there's going to be regulatory bodies who helped with this preventing reference standards or either expected levels of reproducibility and repeatability. This need to be shown in order for a test to be considered valid. So how we implement it and then moving into routine analysis is also regulated as well as the need for performing quality control practices throughout the duration of the instruments life in the lab, and proper reporting of results. So as I mentioned, there's some regulatory bodies who are involved with helping in this situation in order to ensure that. You know, if you have your blood drawn here in San Diego, you're going to get the result, but the same result as if you have your blood drawn in Houston, Texas. And so we have the clinical laboratory standards has to satisfy, as well as the International Organization for Standardization, ISO and ice. Although CLSI has committed strictly to the clinical lab standards, ISO has standards across any, in every topic and discipline, including IOUs to recently for looking up the proper protocol for how to measure the nickel leaching out of nitinol cardiovascular stance, an example. So we want to make sure that any group who's trying to answer the question, how safe is my device is using the same protocols so that the comparison is meaningful and safe for the patient. Really ultimately, excuse me, are the consumers. And here, domestically we have the FDA, CMS and the CDC, or the CLIA being what you'll hear about most often, the Clinical Laboratory Improvement Amendments that was introduced in 19 E. That's used to really ensure that all labs isn't the same protocols. Okay. So just in

terms of validating asset, we want to confirm by examination and provision of objective evidence through a defined process. Process, where this comes from at the particular requirements for specific intended use can be fulfilled. Now, just to give you a view of the CLSI, countless times each day, laboratory tests are used to make critical decisions about patient care. But laboratory testing is not useful in treating patients unless quality and dots are used to ensure accuracy, reliability, and timeliness. For 50 years, CLSI has been setting the global standard for quality and medical laboratory testing, developing an upholding best practices that drive quality test results and improve health care. Clsi has more than 200 published standards, guidelines, and companion products being used by thousands of laboratories in every corner of the world. Clsi work is used by health care providers, drug and medical device manufacturers, and government regulatory agencies to achieve the highest quality results. The benefits are substantial. And in terms of improvement of quality, of accreditation preparedness for those hospitals and healthcare systems for industry in terms of their submissions to regulatory authorities, for government agencies, in terms of the alleviating the need to write new legislation and law and making sure that high-quality laboratory services are provided to healthcare institutions and others. And through its global health partnerships program, CLSI is providing quality and standards training to parts of the world where these resources are not readily available. By partnering with laboratories in these countries and providing laboratory mentors. They can improve the diagnosis of diseases such as HIV, aids, malaria, and tuberculosis, helping to deliver lifesaving treatment. Clsi, improving the quality, safety, and efficiency of medical laboratory testing around the world, leading to better patient care and longer and healthier lives. For more information, please visit our website at CLSI.org. Whoops. So I encourage you to visit these websites and check out what kind of information you can find them yourselves. But now into the statistics, some of the terms we're going to be covering in this part of the lecture include frequency distributions, populations, parameters, statistics, random sampling and probability distributions. And on the right here I show an example that's given in the textbook of a histogram for gamma glued ML transferase enzyme that's measured for looking at the health or disease state of your liver. So it just gives you an idea of how we're going to be using statistics in clinical chemistry. Collecting data from a sample of patients to see what we're going to define as our normal range of values, excuse me, for the population of interests. So getting a little bit more into that here, we define this as a frequency distribution, which is a graphical device for displaying this large set of lab test results. And of course, it is a histogram. Intervals are defined. That makes sense for the measurements that you have and these intervals will get smaller, of course, the more number of samples that you have. So here we have the concentration and press the X axis and the frequency on the y. So if you look at the height of each bar tells you how many observations were made within this interval. And this graph represents a 100 healthy 20 to 29 year old men. The range of values goes from five to 65. And if what we really use these for us when we measure the amount of Juju TNF patient, we want to be able to compare that to the population in general. Where does that person fall in the range? Are they within the normal range? Are they on the outskirts as that indicative of need for medical intervention? Okay. So often we're going to talk about the percentile, what percentile the patient might be at, just as we do with our grades at school. So for example, if you want to know. You have the 90th percentile. If you're in the 90th percentile on your exam, that means you are I have a score that's higher than 90% of everyone else who's taking the test. So in the textbook, they give you this formula for percentile non- parametric approach. But this doesn't give you that percent, percentile, as we're used to thinking about it gives you more the percentile number rank along, along the curve. For looking at the percent. I show you this formula here, which is pounds, the resources on the internet. But I like this one of the best because it shows you if you

a difference. So it's important sometimes to see this coefficient of variance as opposed to just the standard deviation. Okay? So just a note that they define a statistic them as any value that's calculated from the observations in the sample in order to estimate a particular characteristic of a target population. So what is the difference between parameters? Statistics. Hopefully, it's become clear that parameters apply to the whole population, whereas statistics applied to the sample of the population, where we have the parameter being mu and this statistic being x bar or x m, or the average calculated. And this is sample representative of the total population through random selection. Okay, so Gaussian probability distribution then refers to this normal probability that we are talking about already. But in this case, we want to start thinking about the error that's associated with measurements taken in a normal distribution, where the error is defined as your observation minus the average. And that's going to also have a normal distribution. And it's going to obey the central limit theorem or central limit effect, where if you take enough observations, the error will tend to 0. So it is completely characterized by the mean and the variance is often written as N normal normal distribution with mu comma sigma square. And then we talk about the standard Gaussian variable, which is basically going to be a measure of this air divided by the standard deviation, where your population then is described as your distribution is described then as n as 0 comma one, as we approach theme for total population. And, and the probability then of finding any value within two standard deviations of the average is going to be 90.9544. Or this basically is how we define our 95% confidence interval. Okay, so the student t probability distribution as going to be the equivalent of that Z-statistic except now using the sample population again. So we have T is equal to x minus mu divided by the standard deviation. And that's what rights from the textbook. But in most places you'll see it written this way. X minus mu divided by the standard deviation, divided by square root of m, which is the ST divided by square root of n as the standard error actually as opposed to standard deviation. So if a sample size is small, you'll get a greater dispersion with heavier tails and the distribution. And if the sample size is greater than or equal to 30, the student t will approach a Gaussian variable. Again with the central limit theorem so that it converges towards this Gaussian distribution and it'll be tending toward a difference of two. Student t-tests are used for significance testing and for confidence intervals where we often will use that 95% or we said two sigma on the previous page. But in samples they sometimes burgers to alpha. And once you know your t value and you know your degrees of freedom, which is defined as. Minus one. You can look up a p-value for your sample. And the p-value is what's often use to test a hypothesis for whether you're going to accept or reject it. And in our case, this is often going to be looking at whether we accept or reject the fact that our sample mean is going to be equal to the population mean. So for example, or for the mean of another asset. So for example, I just took this from a paper shown here. If you want to look it up later, where they wanted to perform an internal quality control on existing assay. For example, Lydia trophic hormone, hormone. And the null hypothesis is going to be defined as yes, the X mean we're expecting will be the same as the mean we've had in the past, where the significance set to P less than 0.05. So if it falls outside of the 90% confidence interval, basically we're saying that we're going to reject for the manufactures, Yeah, we're going to reject the null hypothesis. But that they want. So they show the results here where for high concentrations of the hormone, they found the average should be well within the acceptable range for accepting the null hypothesis because p is greater than 0.05, so there's no need to. So this appears to be equivalent basically. And here, when they look at the mean value measured with the new essay compared to the predetermined one, it still seems reasonably close together. However, when you go through this statistical analysis, your t-value of 2.3 plus your degrees of freedom, you come out with a p-value of 0.027, which is in fact

less than the threshold that was set. So there is some significant difference there. What is the possibility of rejecting the null hypothesis? So they're gonna have to look into that further and see if this assay is still valid at these concentrations. Now, most biological samples are not parametric. So we're going to be looking at non-parametric samples. And that basically means that a non-normal distribution, they might be skewed or distribution free. So I've already shown you with the G GET samples that that is, that, that is skewed. And here we have a couple graphs showing billy Reuben for population according to gender. So men and women and E because again, we have non-normal distribution skewed to higher concentrations. And so for the descriptors for non- parametric populations, we're going to talk more about the medium instead of the average or the mean. But still, the 50th percentile will give us that. And the lower and upper percentiles, often, we're going to be talking again about the fifth 95th. As again measure of spread. Mann-whitney test can be used. The equivalent of the t-tests for these non-parametric samples. Okay, so we're going to just price here. And actually let me finish this slide and then I'm just going to take a pause and move into the second part of chapter. Okay, so the validity of analytic assays is, is an important part of clinical chemistry. And these are some of the concepts that they just introduced in the chapter that are important for you to be aware of. And we'll see them coming up, popping up here and there throughout the courses we're looking at results from different studies. Okay? Some of the concepts. This is similar to table 2.1, I believe in the textbook. Yes, but I've just added in another column and a couple other terms that you'll find in this chapter as well. So we talk about the terminus of a measurement that's the agreement of the mean to the true value. And this will give us a measure of bias. Accuracy is the agreement of a single measurement with the true value. And this gives us a measure of uncentered certainty. So precision will be a function of repeatability and reproducibility are repeatability is under the same conditions and reproducibility is under conditions such as different time, different operator. And here we're going to be measuring this through the standard deviation or coefficient of variance, is defined by the regulatory bodies that you're going to need at least 20 observations or more to make that meaningful measurement. And we do have to be aware of how calibration or instruments stability can affect these results. Linearity. We often want to be verifying that we're working within the linear range of a relationship between our observable AMS, the concentration. So whether it's looking at absorbance as a function of concentration or an electrochemical result. We want to know that we can, can be looking through linear range of response. That as the concentration increases, that observation will increase in a linear fashion. So that's usually measured through a dilution series. And then we have limit of detection or 11 and quantification to instruments often talk about the limited detection. That's just how low of a concentration they're able to detect. But for laboratory purposes where we're looking at, at a patient value, we really want to know the lower limit of quantification because that's going to be the lowest number on still this linear range of the calibration curve. And that's important if we're trying to quantify the numbers. So and we have sensitivity and specificity, which are important concepts that you also have been hearing about with the coronavirus testing. How sensitive is it has specific is it in the sensitivity is measure of true positives in comparison to the total number of positives measured. So true positives plus false negatives, which means also. And the specificity will be looking at the ability of a test to really discern the target analyte from another interfering substance. So we want to know if it's, if we get a negative result is true negative or is it a false positive? So we're going to look at true negatives compared to all negatives. And that gives us a measure of specificity. Okay? Pause here and then I'm going to resale and the next set of slides in a new recording just for space purposes. Thank you all. And I type and I'll be right back.

you. So if anyone else does now I'm wants to share on the discussion board that will be great. I would also argue some bonus points and otherwise, I found this graph showing how Deming regression differs from linear regression. On the website shown below, can't read here, but you'll be able to see it in the slides. You download them. Okay, so here we have the red line being the Deming regression and then the black line being the linear regression. And the red gives you also the error bounds that are for both x and y. So using these two types of assays you can or statistical, and as you can get a lot of information about comparative, comparing two assets, what ships important? A few other notes that they mentioned and I don't go into all the detail on the regressions were not going to actually be performing any regressions in this chapter. But I want you to be aware of those systems that are used. Okay, so the other notes that are pointed out in the textbook that we should be aware of are outliers. Outliers are often rejected based on some cutoff that is predetermined, whether they're a distance of three or four standard deviations away from the regression line in this case. But you should still investigate their reason in case they occur. Again, correlation coefficient r is defined in the textbook as a relative indicator of the amount of dispersion around the regression line. And they describe the testing for linearity is important. We do make that assumption. We are looking at the concept of linear regressions. So you can perform or runs tests to see if the positive and negative deviations are randomly distributed. In order to affirm that you do can make this linear assumption. And nonparametric regression can also be done using a difference mathematical algorithm. The passing bad luck. Again, this is just so you know that you do have to make these considerations said it differently for non-parametric systems. But in the era we're generally looking at parametric distribution. Okay, so the interpretation of systematic differences, once you've calculated, let's say the intercept and slope from this regression can be done using a t-test where you compare the intercept is 0 and the slope to one. And that will give you more information again about how these error, what the error for, how the error differs from sample to sample from ASA, ASA. Furthermore, they bake a few notes about how this is used in practice. One being that when you're monitoring serial results for patients that are undergoing continued treatment, you need to also introduce error that will arise from Within Subject Variation. So that patient might have some variation from day to day when you're taking these measurements. And there'll be additional piano local error coming into play as well. And furthermore, when looking at assays, you need to be aware of this concept of traceability. So there's going to be a chain of comparisons that occurs, leading all the way up to the top where we have a known reference value, probably determined by these governing bodies. And then you're going to have selected measurements that are performs at the intermediate level, perhaps here, calibrations in your laboratory. And then you'll be doing the routine measurements. And you're going to want to have those compared to statistics run and reported for each of those levels. And then we have this uncertainty concept. They bring up that not only can you measure uncertainty directly from comparisons of assays as we've just been talking about. But it can also be judged indirectly from analysis of intuitive visual error and using the law of error propagation. Okay, so the last part, we're going to talk about diagnostic accuracy of laboratory tests. And here we have, we're going to look at if you have it tests and you are testing just for whether somebody has the presence of a disease or not. Perfect example being coronavirus. While lets go of antibodies in this case, do they have antibiotics or do they not IGM antibodies? So what are the what's the likelihood that you're, well, how accurate your test is will be determined by the following equations. So first you make a table of all the test results and put. The true positives are the permutation does have the disease and you have detected or the duty of the antibiotics we have detected, then you have false positives or they still show a positive result for having antibodies, but the action happened at disease. False negative. Where

you have a negative results but they have actually have the disease and true negative where you see no response and they have not had the duties. So dang that gnostic accuracy is the sum of the true negatives and true positives divided by all of the measurements taken. Specificity will be the true negatives divided by the total number of negatives, whether they're false positives or true negatives. And sensitivity is the true positives divided by the positives. So question, Is it better? I mean, specificity or sensitivity, a better measure of Jim correctly detecting disease. Gave me a second to think about it. What tells you more accurately if you're going to detect the disease. While as sensitivity, because in this case you want to know who has the disease. So looking at the true positives in comparison to the hauled the positive results, we'll give you an indicator of how sensitive test actually will be. And we'll go on to this example in this chapter. Deep vein thrombosis. However, we're going to talk about this next chapter. So deep vein thrombosis can be analyzed by looking at the D-dimer quantity in the patients. And so the D-dimer tests, it's for protein fragment that will come from blood clots dissolved into the body. So this is a blood test that's taking them M and the S. The frequency distribution for patients who don't have deep vein thrombosis looks something like this where they're going to have a low concentration of D Dang Ran they're glad at anytime. And the patients who do have deep vein thrombosis show higher concentration of the D-dimer in their blood. And in this case, it's called a dichotomized index tests because we are going to have a cut- off that we set arbitrarily. So we're going to have to think about where to put this cutoff in order to detect them highest number of true positives and true negatives. In this case, we're more interested in making sure we get the true positives. So they set the cutoff relatively low here, 500 nanograms per mil. And you see now the sensitivity is 97%. So they're accurately detecting the true positives 9, 7% of the time. However, the specificity is only 37%, which means they're going to get a lot of hum. False positives as well. It's the diagnostic accuracy in this case is 50%. This concept of a receiver operating characteristic curve as a means of determining the accuracy of the test is then introduced, where here we black sensitivity versus one minus the specificity at different cutoffs. And in this case, here's our 500 nanogram cutoff. And then we're going to hire Codapps is look at the graphs. I can see the numbers. Here. We had the, yeah, 2133 micrograms per liter. And then down here is the dot for the 5,435 micrograms per liter. So you can see these examples mark in the textbook, or they just show the shifting of the cutoff. And so you can see how many less, how much less sensitive tests is going to be at these higher cutoffs, but more specific. And the higher the area under the curve, the more the area under the curve, the more accurate it is. And basically want to know how much better it is than chance. 150%. Okay? So we're going to come back to this example in a minute, but they then introduce the posterior probabilities and odds ratios. And I'm including P-values here because it's a really important measure that's used in statistics. One looking at whether or not a new essay or test is statistically relevant or helping. Showing that as that it's improving what you're currently doing. And with statistical, what's the word I'm looking for? Significance. Statistical significance. Okay. So anyway, the posterior disease probability are often just stated. The desired probability is similar to sensitivity, but instead of total positives divided by all the positives, now we have total positives divided by total positives and false positives. So we want to know the probability of having a positive result based on the fraction of total pastors had positive results. And they also show similarly the probability for him than negative. I'm result oz ratio, on the other hand, is an alternative way of expressing probability, where we show that an outcome will occur given a particular exposure in relation to not occurring. So it's formulation here, probability over one minus probability to, for example, if the probability of getting a true positive is 80%, the odds ratio is four. And anything above one means you have a higher odds ratio. It's a higher chance is going to happen. And if is equal to one, then

at this point, we're finished with this chapter on statistics. Perhaps your mind has been blown and whatnot. Another way. But I'm glad that we've gotten through this part of the course and we're going to talk a little bit about statistics are still in terms of diagnostic accuracy, sensitivity, and selectivity all the way through the course. But hopefully this will give me a good base for understanding that and moving on to some other interesting topics Lecture 3 Transcript (Ch. 5) Okay, welcome to the third lecture in this week on Chapter five, reference values. I know we talked about this in the previous chapter and we're going to talk about them many, many times because they are very central to click chemistry. And I think you're going to find that there's quite a lot of review in this chapter from the things we discussed in chapter two. So hopefully that will be helpful since there was so much information in the previous chapter. And this one, I think you'll find it to be shorter and easier to digest. Now. In even the beginning of the chapter where we discuss how is a diagnosis made? Something reject about in chapter two as well when we're looking at workflow, the end. So in practice, a decision is made, the diagnosis is made, and a medical decision is made based on all the collected data. That includes the medical interview, physical exam, as well as the medical laboratory test results. And these results are always compared with reference data. So the decision-making is done by comparison. And the data really would be useless without something to compare to. So I just found, hey, we really need these reference values and aren't and make any, draw any conclusion about what's going on. Now, the reference values really needs to be from representative populations. And those are going to also need to draw from all types of people, including healthy and disease, hospitalized and amulets R3 patients. Okay, so to establish a reference value, it's been determined that there are some mandatory conditions in order for it to be considered valid. So first of all, groups of reference individuals should be clearly defined. The patient that's being examined should really resemble the reference individuals in our respects other than test aspects. So for looking at a diabetic patient, we want the reference individuals to be diabetic patients as well for trying to figure out what's a normal range or acceptable range or a certain value. Now, the samples should be obtained and process similarly for the test of the patient that is in question as to how the information was obtained for the reference population. And it must be well documented so that we can refer back to that and make sure that indeed they were done in a similar fashion. The analysis quantities must also be the same. And the results from standardized methods. And the results must be obtained from standardized methods. And Paul will always using sufficient quality control. And then of course, the known clinical sensitivity, specificity, and prevalence in the population should be included. And that will be helpful in the decision-making as well. Okay, so who should be included or excluded from this reference population? We often talk about selection criteria for patients. In terms of inclusion or exclusion criteria. A random sample as desired where there's, because otherwise there's potential for bias. But the random sample might come from a certain population. So we can partition by a number of different conditions. Sex, sex is one we even looked at in the last lecture. We're see different reference frequency distribution SUS by gender for bilirubin and for G2D ci GET reviewed, revisit again in that chapter. And it could be by blood type or ethnicity or other criteria as well. It's on table 5.1 in the textbook. They give some examples of partitioning criteria. And that means not that we're going to exclude it, but that we're just going to separate it and use it for the appropriates patient comparison. Now on the other hand, exclusion criteria, which

includes some of these same criteria, would be to remove them from the population, reference population in order to get a better view of where we think our patients value should fall. So this could include pregnancy, drug use, recent hospitalization, transfusion, or any number of exclusion criteria listed here in table 5.1. Now, an important consideration when we are looking at that reference values compared to the patient diagnostic test, is that we want to be sure that in addition to all the things you just mentioned that the pre analytical process is also standardized. We want to minimize any bias or biological noise. And so we want to make sure that the individuals are prepared for the test in the same way. So for example, if we know we want to eliminate the interference of just sensible drug use, then we ask them to refrain from doing so for two days before the test. It also matters often how the collection itself is made. So body posture will actually influence some of the non-feasible analytes such as serum element. So you want to have your patient sitting not lying down, for example, in order to ensure that you're getting the similar results from different patients. And or even if you're Jang intersubject record within the same subject, subject based comparisons than you especially want to make sure that your collection procedures are the same. And how we handle two-sample prior to analysis is also critical. So. Once we're ready to begin to tearing the reference values, we need to keep in mind that method of analysis that we're going to use. And we'll be sure that that is standardize too good an equipment. The thing is the reagents that we're using, the calibration of the equipment, raw data at the calculations we used to determine the value. These are all important things to ensure. We have them or the same in both. The reference value that are obtained as well as the diagnostic test for the patient. And quality control, as well as really critical. And we now know that the labs are using some kind of quality control, either or both internal and external, comparing to other labs. And internal, like if you change a reagent or new reagent, you still getting the same results. And so that is also related to reliability, where we're looking at a statistical analysis to show how reproducible and how repeatable our measurements are and who I ensure that these are high-quality in order to be confident and any result that we're obtaining. Ok, so once we collect the data, then what? Well, as I said earlier, we may partition the data in order to be more representative of the distribution that makes sense for the patient or comparing the data too. And we want to collect a sufficient amount for the partition. And if we can't combine them, if it makes sense. Because the more the higher the value, the better your information as I've represented the true population as we learned in the last chapter. So you can also inspect the distribution visually at this point to look to see if it's skewed or normal or bimodal or if there's any visual is visible visual key issues to the reference limits themselves are outliers. So here's an example of, again, we're back to the gamma gluten will transferase as we're looking at in the previous chapter. But there's different frequency distribution. So you can see is still not normal. It's skewed to the upper concentrations and there is a data point way up here. So I think that's an outlier. And then if so, we want to be able to test to see if it truly is or not. So we see that deviate significantly, but we'll, but we have to be careful because and we see that skewed this direction anyway. So it could be a true value. But if we use the Dixon read range tests that the textbook introduces, where we look at the quotient between the difference of the two highest values are the two lowest values divided by the total range. And then looked to see if this quotient as This ratio is greater than or less than as certain credit off, in this case 33%. And enlist scenario. We see that indeed when we make this calculation where the difference between these two values, 74 minus 50 is 24. And put that over the spread of the data, 60 gives us 0.35. Indeed, it is higher and that would give us a reason to reject this outlier and just treat this as these sample population. Ok, so once we determine the reference limits, well, we, sorry, so last way to turn reference limits just by removing the outliers, we can't really calculate with this interval is going to be as shown here and described in the

the interval being two standard deviations from the mean. And they do give an example in a textbook. A textbook I'm not gonna go into details on. I just want to point out that for the parametric, it's really highlighted the importance of testing for goodness of fit. So you want to be sure that the cumulative frequency when plotted on Gaussian probability paper is linear. And art, be sure that you can use the parametric statistical analysis. So if you find that it's not as in the graph here, near linear fit, doesn't seem that linear. There are some transformation that will allow you to get marmot linear fit and must be able to use your parametric statistical analysis. So for example, sometimes if it looks as such, we might try either Jin, the log y equals log of the concentration or square root of the concentration. And in this case they show same data plotted as cumulative frequency versus log of GET. And now we do see Marvel linear relationship. So table 5.4 shows this just to show you that if He is the non-parametric method to determine the reference intervals for the same data set. Three different methods. Ok, so the nonparametric gives us a lower limit of seven, upper limit of 47 with confidence intervals Sean and brackets. And we see there's one value below the lower limit and two values above the upper limit for the parametric untransformed. So that will be this plot up here. We get a lower limit of 0, but it's bounded by a negative number and an upper limit of 36. But here we see that there's no values below the lower element, but there's seven values above the upper limit. And but on the other hand, if we use the parametric transfer form data as for according to this grabbed here, then we see the limits are more in agreement with the nonparametric. We have one value below and six above. Now. So even though we have used a system is linear and still appears that the best approach for this data set would be the nonparametric. Where are, where we're not so skewed to lower, to the lower end of the concentration values. Okay? So how do we use these reference values? Well, as we've been talking about this lecture and the previous one, we're going to interpret our medical laboratory data comparing patient's values with the reference value. And as we've also seen before, flag it as high or low if it's outside of the interval. So this is just another, for example, a test from a patient. And here we see muster the values are falling within that range except orange. But glucose here is shown to be low and so it's been flags and that allows the doctor to follow up on that. Necessary. Okay. So note about subject base versus population based. We need to be aware that a change in an individual that may still be within a reference interval, this thought to be associated with health may not be normal for that patient. So figure 5.3 shows us, let's say here this patient a, and you get a measurement for whatever we're measuring, let's say Juju t of b, that would fall within the normal range where it wouldn't raise a flag. However, if this is your normal than it is outside of the normal range, is indicative of something important. And so we want to be sure that we're keeping an eye on these data as well. If we have access to a comparison to the subject base reference values. And this, another example here shown on the right, looking at iGEM and with as a function of, of time basically of when they've been monitored for different patients. So this just shows that the intra individual versus the inter-individual variability is quite different. So this patient is going to be well outside the range of the others, but their value within one comparing themselves doesn't vary very much. So it's important to be aware of these things. Transferability and other key concepts where we need to know are these populations really comparable? Or they're standardized protocols from specimen collection to analytical purpose that are being carried out as their common calibration that we can use to be sure that we are actually able to compare the results from labs. And is there an external body control and affect? So sometimes we'll actually use reference value so that we haven't determined ourselves back probably quite often. And these could come from manufacturers inserts. So sometimes the company who makes the tests will provide the reference value to could've come from peer reviewed publications. Scientists have been studying this and found a good value, then they will

use reference range that they will use. And it could come from multicenter. Trials so that you are really increasing your population and correcting for any variability from lab to lab. To verify the verification of the transfer. Of course, there is the CLSI, who's the governing body we mentioned before, where they can actually provide the internal calibration. But they also give guidances on how to do so. For example, at the minimum of 20 reference value measurements and less than two that fall outside of the range that you're looking at. Okay? And then we have another note about the selection of the reference values. So labs are really responsible and do take the lead on providing reference limits for the tests. But and is multi disciplinary group that would decide including clinicians, Anne's clinical laboratorians. So it's important to understand these so that you can actually weigh in on the conversations about what reference values make sense in your experience. Okay, so this page really should look familiar because we're just looking at the sensitivity and specificity again, where we know we have our true positives, true negatives, false negatives, false positives for patients with disease without the disease, et cetera. And the formulas are just shown in the table here where they weren't before for sensitivity specificity and then calculates the total negatives, positives, and gives the predicted value we did tackle or the predictive value before. But let me just remind you that this is basically if you get a positive result, what is the probability that the patient is actually positive? So that's predictive value for positive test, semi negative tests. Now, concept we didn't really talk about in the last chapter is prevalence. Which just means how much of the population actually would be expected to have the disease. Because this, it turns that does affect that predictive value. And really that's important for knowing how valuable a test really is. So these two figures show same disease, but with different prevalence. So let's say the prevalence is 50%. And we're told that our instrument and has a sensitivity of 95% and a specificity of 90%. And that's pretty good. So we find that we will, our predictive value for patients who test positive to actually be positive is 90%. And F for patients who test negative and RNA good, that predicted value is 95%. So we're doing a pretty good job predicting here out once the disease prevalence goes down to only 5%. That's really changes. We still have sensitivity 95% and a specificity of 90% But now, when we put a plug in the numbers, we get a predicted value for the positive as much reduced. Actually, they wrote 5% here, but I think that was just an air of copy error. The actual value, if you calculate yourself, you'll find is 33%. And that's also in the figure caption. But regardless, 33% and still much lower than the 9% here. And the negative prediction adsorption, which is nice, but often we want this number to be higher to. Okay? So I'm just going to show here a couple of examples of how this is important in our currents fight against the coronavirus. So here is a graph from the literature from this paper shown here that is highlighting are showing out the cut-off is being chosen and trying to determine the probability of having a false positive or false negative tests for the sparse code e2. So, so this graph is discussing the problem with false negative tests. So if we have PCR tests for the coronavirus, that's not very sensitive than we are likely to get more false negatives. And but this is actually graphing AMS, the chance of infection if we have a negative result when visiting somebody else versus the chance of infection. If you haven't had a test and you don't know if you could be an asymptomatic carrier. So if you have the ability to get a very accurate test and any year, more confident and you're negative test result. It means that you have more of an opportunity to spend time with people who might be worry about affecting before you have to be concerned that you might, in fact, while I was confusing, but anyways, this is just an example of how these statistics are being used currently right now to try and better understand how important social distancing as for example, and how effective it is and how important it is to have really accurate testing. Now, you also probably hearing a lot about the antibody test and it's the news allowed as well. So here's just an example of one of the commercial assays that are out there today, the avid IgG essay. And it has been shown

much of our body white is actually? Well, it's a rant about 8, 8% percent of our body weight is blood. From miles. We could tell her at about five to six liters of blood. And for females it's around about four to five liters of blood. And the temperature blood in our body is about 38 degrees and blood is the primary way of shifting hate around the body. If it gets too hot, then the blood vessels dilate. We released hate. If it's too cold, blood vessels constrict. We hold on to that hate. And what type of tissue is blood? Remember this full tissues of the body? Nervous epithelial, connective and muscle. Blood is connective tissue. Remember, connective tissue is a whole bunch of bonds and. And so bargain is connective tissue. You've got cartilages connective tissue, and you've got blood is connective tissue. And connective tissues are made up of cells, gels, and fibers. And it's the concentration of fibers in the types of gels that depend on the viscosity or hardness of the connective tissue is obviously few farmers in blood and therefore it's a liquid. So what I've drawn appears a chub. I've taken my bladder, popped it in the tube. I put this tube in a centrifuge, and I've spun that centrifuge around and what that does is overtime. It separates out the components, the major components of blood according to weight and size and what we get. And the biggest and heaviest things down the bottom. And we get the largest and smallest things up the top. Alright, there's three layers we need to talk about, right? 123. So the first layer is the layer at the top that we're going to focus on to begin with. And that is what we call blood plasma. And the blood plasma consists basically it's 55% of your entire blood volume is blood plasma. And blood pleasure is made up of three main things that you should know. It's made up of water, proteins and solutes. So out of this 55%, you're going to find that 92% of plasma is water sucked. If most of the blood is plasma, Most of the plasma is water. Most of our blood is water. Alright? Proteins make up a random at 7% and you're going to find that solutes is less than 1%. Now, with the proteins has three main types of proteins in blood that you should either somebody saying well, but this three mind that you should be aware of. So these three proteins, albumen. Let's first focus on album and then I'll talk about the. So album in is an important, most abundant protein in our blood. It does a couple of things rot. So one, its carry out will transport protein. And what it usually carries around are substances that are lipid soluble. So if it is lipid soluble, it doesn't like water. Most of the blood is water, so it doesn't want to be in the water, but it still needs to be transported. So bonds to albumin and lipid soluble substances, rock. Lipid soluble. They convey lipid soluble drugs for example. Or they could be lipid soluble hormones. And the lipid soluble hormones include things like steroid hormones, right? The second topic prior to, and you should know, the globulin and the global ones play a big role in immune function and clotting. Let's just run immune. All One thing I forgot about the albumen because I said one is obviously another point. He for element is that it is the most important protein when it comes to maintaining osmotic pressure. What's osmotic pressure? So remember, osmosis is the movement of water towards an era of high solute concentration. So if you think of a blood vessel, for example, a capillary, capillaries have holes in them and they fade. The tissues outside of that blood vessel is some cells that need to be fed. Then it's the bay oxygen and nutrients, for example, that need to come out to fade those tissues, but at the same time what it comes out. So throughout the day your capillaries are constantly leaking fluid. Now, over an entire day, if none of that flowed was reclaimed, would lose most of our blood volume for May five to six liters worth of blood volume just in the periphery or in the interstitium between the blood and the tissues. That's where my fluid will go on. Blood pressure will go down when bigots situation. So I need a y of reclaiming that fluid back in. And the major why I'm doing this is albumin. Protein that sits in the blood. Protein has a negative charge. It loves pulling water towards it. And that's called maintaining osmotic pressure album. And just like globules rotten, just like fibrinogen, which is going to be the third prototype. All made in the liver. So the liver isn't doing too well. You might not produce enough album and you

may not maintain osmotic pressure and fluid. My remind late debt and this is edema or I'm sorry, really important clinical link there. So he said globules are important with immune function, but there are also important in cloning. And then the third protein is fibrinogen. And fibrinogen is an inactive protein that needs to be activated into fibrillin rot. And it's involved including as well in the clotting cascade. Perfect. But what about Soviets? Will top of Soviet still we have the Soviets are going to be things like ions. And ions are sodium, potassium, magnesium, calcium chloride, things like that. Nutrients. There's nutrients may be glucose or amino acids, fatty acids, for example. Gases. Gases like oxygen and carbon dioxide and nitrogen. And waste. Uric acid or ammonia for example. Weiss metabolic wastes is what we're referring to. So you're gonna find that plasma makes up the most of in Thai blood fit 5% water proteins solids, and these are the components of that. Next part is this part here. So this part here is the smallest component of our blood called the buffy coat. It is this white buffy layout. If you put it in a centrifuge, it makes up less than 1% of the entire blood volume, and it is made up of leukocytes and thrombocytes, which are white blood cells and platelets. So leukocytes, the student leukocytes first. So for the leukocytes is around about 10 thousand cells per mil, right? And Lucas science, like I said, I'll also known as white blood cells. Lu Kai means one sought main cell phone, different types of leukocytes. Remember, the mnemonic, never let monkeys eat bananas, never let monkeys ate bananas. There's your mnemonic. Neutrophils. Lymphocytes. Mano science, eosinophils, basophils. And it's also guys in abundance, most abundant to laced abundance. So most of our blood cells are neutrophils really important in what's it called when you get damage, vascularized tissue, inflammation. So that's Lucas on to the other one was thrombocytes, which are platelets. That most cell fragments then cells themselves from mega carriers thoughts. But like I said, platelets. And you have a rant about 300 thousand cells per mil, right? So there's more platelets in number than there are white blood cells. And platelets play really important rolling clotting. So these leukocytes, white blood cells are therefore immune function, right? So you get TE base cells and all these other cells that have really important nuclei as well. So we'll talk about that in a future video. And platelets which are involved in the clotting cascade. All right, the last one in the bottom is eosinophils, Saris or erythrocytes. What am I talking about? Erythrocytes, which means red cells. Lucas Antoine cells, erythrocytes, red cells. So the ABI sees red blood cells and weave around about 5 million per minute, one of the most abundant cells in the entire body, and what they do is they carry gases, right? They carry oxygen, carbon dioxide, really important. Red blood cells are filled with hemoglobin that carry oxygen. Okay, so when we take blood and we spin it down and we measured this, the percentage of this thumbnails is around about 44%. For women, the red blood cell percentages around about 40%. And this is also called El hematocrit, right? So measuring have Maddie crit is simply measuring the red blood cell percentage of whole blood miles around about 44% females are NBA 40% plus or minus a number of percentage. Now here's the thing. The reason why we do this is if it goes too low, it might be an indication of anemia. Not enough red blood cells. If it goes too high, maybe an indication of poly soothe female. And these will be the topics of future videos. So as we look at the hum, adequate or blood components, three major types. Plasma effect of 5%, Buffy count less than 1%, and erythrocytes are at about 44%. So hopefully that helps looking at blood components. Okay. I hope you enjoyed listening to him breaking down what's in our blood as much as I found it useful. And they have a lot of other videos that you might find helpful as we go through the course material. Obviously, definitely more well versed in the anatomy and physiology parts of what is important to clinical chemistry. Next slide. Okay, so plug collection obviously is an important mass spec them and as you may know, it's. Verges phlebotomy. One of the most unusual words in this course, I think phlebotomist has such a funny thing, but it's a funny name for a very important job. Ok, So usually it's venous blood that's going to be