Deciphering Sources of Variability in Clinical Pathology

Abstract

The objectives of this session were to explore causes of variability in clinical pathology data due to preanalytical and analytical variables as well as study design and other procedures that occur in toxicity testing studies. The presenters highlighted challenges associated with such variability in differentiating test article–related effects from the effects of experimental procedures and its impact on overall data interpretation. These presentations focused on preanalytical and analytical variables and study design–related factors and their influence on clinical pathology data, and the importance of various factors that influence data interpretation including statistical analysis and reference intervals. Overall, these presentations touched upon potential effect of many variables on clinical pathology parameters, including animal physiology, sample collection process, specimen handling and analysis, study design, and some discussion points on how to manage those variables to ensure accurate interpretation of clinical pathology data in toxicity studies. This article is a brief synopsis of presentations given in a session entitled “Deciphering Sources of Variability in Clinical Pathology—It’s Not Just about the Numbers” that occurred at the 35th Annual Symposium of the Society of Toxicologic Pathology in San Diego, California.

Clinical pathology testing is routinely conducted in toxicity studies during the drug development process. The instruments used in clinical pathology testing usually go through robust validation protocols, and procedures are guided by “Good Laboratory Practices” and other guidance documents. However, even with these efforts in place, variability is often encountered in clinical pathology data that can be broadly attributed to preanalytical, analytical, and postanalytical factors. Some of these variations may be expected, such as those related to study procedures or some preanalytical variables (e.g., order of collection), whereas others may be unexpected (instrument error) and only identified upon close monitoring/scrutiny of the data by the laboratory staff. This session explored the causes of variability in clinical pathology data and assessed the impact on overall data interpretation and determination of test article–related effects. The first two presentations focused on preanalytical (Everds) and analytical (Schultze and Irizarry) variables and their influence on clinical pathology data. The third presentation (Aulbach/Provencher/Tripathi) highlighted the influence of study design–related factors on clinical pathology results. The last presentation (Hall) emphasized the importance of various factors that influence data interpretation including statistics and reference intervals and appropriate use of qualifiers to describe magnitude of changes in various parameters in clinical pathology report. A brief summary of these sessions is provided below.

Preanalytical Considerations for Clinical Pathology Data

Everds

Dr. Everds discussed the impact of various preanalytical factors on clinical pathology parameters routinely evaluated in nonclinical safety or toxicity studies. Preanalytical variables can complicate data interpretation making it more challenging to identify test article–related effects. Some of these variables may be controlled with careful attention and planning, such as fasting status, restraint and anesthesia, timing and order of clinical pathology collections, and phlebotomy technique. Knowing how these procedures impact test results is important in minimizing preanalytical variability in toxicity studies.

For human laboratory medicine, hemolysis of clinical chemistry samples and clotting of hematology samples are the two most common preanalytical issues; not surprisingly, these are also frequent complications in toxicity studies. However, unlike human laboratory medicine, other preanalytical factors such as age, diet, and environmental conditions are usually well controlled in toxicity studies. The main sources of preanalytical variability in toxicity studies are attributed to study procedures, husbandry practices, restraint and anesthesia, venipuncture, and sample processing. Fasting status can impact not only blood glucose concentration and parameters related to energy metabolism but also other tests. For example, nonfasted rats can have high creatine kinase protein and parvalbumin concentrations, marker for muscle injury, compared to rats that have been fasted overnight. The timing of clinical pathology sample collection is another factor that impacts clinical pathology data; this variable includes not only the time of day but also timing relative to other procedures. Diurnal patterns are well known for several clinical pathology parameters including leukocytes and hormones but can also be apparent in some other parameters, such as triglycerides in mice (Pan et al. 2010). The effects of diurnal rhythms are usually more evident in rodents compared to large animal species. Study procedures involving animal transfer or handling/restraint of animals may show evidence of stress in clinical pathology data, such as increased cortisol levels and decreased lymphocyte count (Everds et al. 2013). Influence of anesthesia on clinical pathology data is also well documented, especially in rodents (Deckardt et al. 2007). For nonhuman primates, the effect of intramuscular ketamine anesthesia is minimal on clinical pathology parameters collected immediately after injection. However, subsequent to ketamine injection, skeletal muscle parameters can be increased for up to 5 days (Davy et al. 1987). Habituation of rodents and positive reinforcement (such as voluntary blood collection, paired housing) in nonhuman primates are a few ways to reduce variability in clinical pathology data. The order of sample collection and analysis has potential to introduce preanalytical variability (as well as analytical variability in some parameters, such as electrolytes). Thus, sample collection and analysis in randomized order is recommended to avoid bias in certain groups. The site of blood collection and total blood volume being collected also influence clinical pathology data, especially hematology. In general, samples from more peripheral vessels (e.g., tail vein) have higher red cell mass, white blood cell counts, and platelet counts than those from more central sites (e.g., jugular or cardiac). The difference between sites tends to be more pronounced in rodents compared with large animal species (Doeing et al. 2003; Nemzek et al. 2001). The maximum blood volume to be collected is generally based on the Institutional Animal Care and Use Committee guidance that are designed to support animal welfare. However, collecting maximum allowable volume can still impact clinical pathology data making it difficult to detect test article–related effects or reduce the ability of animals to adapt to test article–related effects. For coagulation testing, the blood-to-anticoagulant ratio is critical; collecting smaller blood volumes than required can cause prolongation of clotting times. Sample storage prior to analysis can impact test results as well. For hematology, slides should be prepared and fixed or stained immediately after collection. Analysis is generally recommended within 6 hr of collection, but samples can be refrigerated overnight if necessary. Coagulation and clinical chemistry analysis should be analyzed the same day or samples should be frozen for future analysis. Long-term sample stability is usually best at −80°C.

Recognizing and Reducing Analytical Errors and Sources of Variation in Clinical Pathology Data in Toxicologic Pathology Studies

Schultze and Irizarry

Drs. Schultze and Irizarry pointed out several factors that may introduce analytical variation/errors in clinical pathology data and emphasized that veterinary clinical pathologists, with their education and training, are uniquely qualified to assist in investigation of unexpected results. Although causes of unexpected results and errors are more commonly attributed to preanalytical and postanalytical factors, true analytical errors may occur in a relatively small proportion of cases. For analytical errors, instrument malfunction and operator error were cited as two major causes (Hooijberg et al. 2012). Analytical errors have been a cause of major concern in human medicine, particularly for immunoassays, because erroneous results can adversely affect patient care (Plebani 2006, 2007, 2009). Similar concerns have been expressed for animal species due to increasing use of immunoassays in drug development. One example cited for analytical error mentioned decreased serum glucose concentration due to delay by clinical pathology laboratory staff in separation of serum from the clot resulting in increased glucose metabolism by the erythrocytes. An estimated 10% decrease in serum glucose can occur per hour due to glucose metabolism by the erythrocytes when samples are left at room temperature without separating serum from the clot (Evans and Duncan 2003). Another example was for assay interference in which increased serum creatinine concentration occurred due to positive interference caused by the test article with the creatinine assay (Jaffe’s method). This interference was not observed with enzymatic creatinine assay which provided accurate results, so enzymatic method may be considered where assay interference is suspected with commonly used Jaffe’s method. A third example pointed out instrument malfunction where unusually high platelet and low erythrocyte counts were produced by the hematology analyzer due to presence of small blood clot in the instrument. Drs. Schultze and Irizarry emphasized the importance of appropriate validation of laboratory instruments and procedures for generating accurate data. At the same time, they stressed the importance of close data monitoring by the technical staff to avoid such instrument errors, even in the laboratories that have gone through rigorous validation procedures. When selecting a laboratory for outsourcing of clinical pathology work, they also suggested paying attention to training and qualifications of the laboratory staff and favor using the laboratories in which the staff has proper training and appropriate credentials.

Influence of Study Design Variables on Clinical Pathology Data

Aulbach/Provencher/Tripathi

Dr. Aulbach et al. explored study design variables that can impact clinical pathology test results and confound data interpretation making it difficult to identify test article–related changes and differentiate them from those attributed to study procedures. They asserted that such variables should be minimized in toxicity studies. Some of the examples cited included changes related to species and test system, animal age, husbandry practices, study procedures such as fasting, routes of dose administration, and factors related to class and mechanism of the test article and vehicle effects. The example for effect of animal species and test system pointed out distinct patterns in certain clinical pathology parameters that occur with age and rapid growth phase. These changes are readily apparent in adolescent animals such as rats and dogs and become more discernible in longer-term studies. Some frequently identified trends in growing animals included steadily increasing red cell mass (i.e., red blood cell count, hemoglobin concentration, and hematocrit) and serum albumin concentration and decreasing alkaline phosphatase (ALP) activity and white blood cell counts (primarily due to decreasing neutrophil and/or lymphocyte counts) with maturity of animals (Provencher-Bolliger and Everds 2010). Decrease in ALP activity is a prominent age-related change, generally attributed to decline in bone growth with maturity (Rosset et al. 2012). Clinical pathology changes related to old age are generally evident in carcinogenicity studies primarily attributed to naturally occurring diseases and opportunistic diseases or conditions associated with general morbidity. However, their impact on clinical pathology data interpretation is usually minimal, and full clinical pathology assessment is not recommended for carcinogenicity studies (Young et al. 2011). Among husbandry practices, animal transport and acclimatization, housing, and fasting are known factors that influence clinical pathology results. As an example, they presented data from dogs, suggesting that at least a 1- to 2-week acclimation may be necessary following the transport to allow acclimatization of animals and return to homeostasis (Obernier et al. 2006; Ochi et al. 2016). With inadequate acclimatization following transport, stress-related patterns are frequently encountered in leukocyte differential cell count, referred to as “stress leukogram,” which generally consists of increased total leukocyte count and absolute neutrophil counts and decreased absolute lymphocyte and eosinophil counts. Dogs also have a corticosteroid-induced isoenzyme of ALP which may be increased after a stressful event (Stockham and Scott 2008). Benefits of group housing were pointed out for most animal species that generally results in less distress (Everds et al. 2013). Although minor differences have been reported in clinical pathology data of individually housed versus group-housed monkeys, these differences generally do not impact the overall interpretation in toxicity studies (Voyer et al. 2016). Fasting is usually recommended to reduce variability in hematology and clinical chemistry data (Weingand et al. 1992; Hall 2001). Route of administration was cited as another study design–related factor where intravenous (IV), subcutaneous (SQ), and intramuscular (IM) routes have more potential to influence clinical pathology data than oral administration. Clinical pathology markers of inflammation (such as increases in total leukocyte count and concentrations of acute phase proteins and cytokines as well as serum protein changes of decreased albumin and/or increased globulin concentrations) are frequently observed that are generally related to indwelling catheters for IV and muscle/tissue injury for SQ and IM administration. Similar changes of inflammation or skeletal muscle/tissue injury are also observed with studies involving surgical procedures (such as wound healing, stent placement, etc.). With IV administration, presenters cautioned about the potential for “dilutional effect” that can occur due to fluid contamination if clinical pathology samples are drawn from the infusion catheter during or shortly after the infusion procedure. Additional factors to be considered are those based on the class or mechanism of the test article including dose–response, exposure and clearance of the test article, and expected pharmacologic/pharmacodynamic effects. Unlike small molecules, lack of a dose–response is a common occurrence for large molecule/biologics due to target–receptor interaction. With biologics, lack of effects in individual animals is also common due to development of antidrug antibodies, which may lead to immune complex formation/deposition. Toxic effects of vehicle formulations should also be considered depending on the nature of vehicle. The presenters cautioned against some commonly utilized practices (e.g., collecting samples in group order) or study designs (e.g., excessive phlebotomy) that limit our ability to provide accurate data interpretation.

Clinical Pathology Data Interpretation and Description: Statistics, Reference Intervals, and Severity

Hall

Dr. Hall’s presentation focused on interpretation of clinical pathology data for toxicity studies and various factors that should be considered. He emphasized the importance of gathering all relevant information, including review of protocol, any study procedures that may have impacted clinical pathology data, and knowledge of other study data that can help with clinical pathology data interpretation (such as clinical observations, food consumption, body weight, exposure data, histopathology findings, etc.). Additionally, he mentioned the importance of reviewing individual data tables, along with the summary tables, to get a good impression of inter- and intraanimal variability that can help with identifying test article–related effects and dismissing changes attributed to biologic/physiologic variation. For identifying test article–related effects, weight of evidence approach should be used where several factors are taken into consideration such as magnitude of change, presence of a dose–response, consistency over time and/or between sexes, and correlative changes in clinical pathology data and other study data. Statistical analysis should be used judiciously as one of the tools for such decision-making but not as the sole criteria for identifying test article–related effects. Dr. Hall also emphasized that reference intervals have limited value for interpreting data in toxicity studies because appropriate partitioning factors are generally not taken into account. Additionally, the animals used for reference intervals may not accurately reflect the study procedures and environmental conditions for a given study which can greatly influence clinical pathology data. Concurrent controls are usually considered more relevant for interpreting study data (Weingand et al. 1992) because they are age-matched animals and exposed to similar study procedures and environmental conditions as the animals dosed with the test article. Another topic was severity scores, where he clarified that commonly used severity modifiers (such as minimal, mild, moderate, marked, etc.) are based on subjective assessment. A strict numerical cutoff cannot be applied for these severity scores because each clinical pathology parameter is different, and a certain magnitude of change can mean different things for different parameters. With this premise, he cautioned against use of such magnitude alone (e.g., percentage or fold change calculations without descriptive modifiers, minimal, mild, etc.), as that can be misleading to some reviewers who may get caught in the apparently largest change in magnitude without understanding its biologic or toxicologic importance. For example, a 500% increase in basophil count may seem very prominent but does not have much toxicologic relevance, whereas a 20% decrease in sodium will be considered a “marked” change in this parameter and toxicologically important. For clinical pathology test results, determination of adversity is only possible in instances when an analyte may be critical for health and considered adverse because of the mechanism or severity of the change. However, in majority of cases, clinical pathology data are considered markers of an adverse effect where they may be associated with findings that are considered adverse based on another end point (such as histopathology).