Comparison of IDEXX SediVue Dx® urine sediment analyzer to manual microscopy for detection of casts in canine urine

Abstract Background Detection of urinary casts is difficult due to their intermittent presence and deterioration in urine samples. Objective To compare the performance of the IDEXX SediVue Dx® Urine Sediment Analyzer (SediVue) with manual microscopy for the detection of urinary casts. We hypothesized that the SediVue analyzer would perform similarly to manual microscopy in cast detection. Animals Four hundred forty‐three samples from 420 dogs from a hospital population. Methods This is a prospective, cross‐sectional study. For SediVue analysis (software version [SW] 1.0.1.3), uncentrifuged urine was pipetted into a disposable cartridge. Seventy images were captured and processed by an onboard algorithm. For manual microscopy, urine was centrifuged to obtain sediment. Any cast identified by either method was considered a positive result (>0/low‐power field [LPF]). SediVue images were evaluated if casts were detected by either methodology. A revised sensitivity and specificity were calculated after image review and when using a threshold of >1 cast/LPF. Results The sensitivity of the SediVue analysis for the detection of urinary casts was 53.7% (43.85%‐63.35%), and specificity was 86.0% (81.78%‐89.51%). After image review, the revised sensitivity/specificity was 52.0% (42.89%‐61.02%) and 90.6% (86.81%‐93.54%), respectively. When using a more clinically relevant threshold of >1/LPF, the sensitivity was 52.6% (35.82%‐69.02%) and specificity was 99.3% (97.85%‐99.85%). Conclusions and Clinical Importance The SediVue provides moderate agreement to manual methodology for detection of casts in urine.


| INTRODUCTION
Urinary casts are concretions of mucoprotein or cellular elements (intact or breakdown products) that form in the distal renal tubular lumens and that are shed intermittently in the urine. 1 Casts are often observed in humans with concentrated urine, acidic urine, or under conditions of low urine flow but can also be observed in urine samples from humans with primary renal disease or systemic disease that causes secondary renal disease. 2 In the absence of a disease process, 1 to 2 casts/low-power field (LPF/10× microscope objective) can be seen in clinically normal dogs and cats. 3 Casts provide valuable clinical information about kidney health at low cost and with minimal invasiveness, although their detection can be problematic. 4 Urinary casts are prone to physical deterioration, especially in alkaline, dilute, or stored urine. 3,5 Immediate (<1 hour after collection) analysis of fresh urine specimens is critical for their discovery and evaluation, yet routine urine assessment is often delayed or sent to reference laboratories for analysis. Moreover, there is high interobserver variability and inaccuracy with regard to identifying casts in human urine sediment. 6,7 It has been speculated unfamiliarity, and lack of training in urine sediment examination by veterinary technicians and veterinarians might decrease accurate identification of formed elements including casts. 8 For instance, manipulation of light alignment on the microscope is critical for the identification of formed elements, including casts, which might not always be implemented in veterinary practice.
Several instruments have been introduced for use in the clinical evaluation of human urine in an attempt to automate urine sediment examination and minimize these challenges. These laboratory analyzers bypass the issues that might occur due to sample shipment and offer the potential of increased intra-assay precision, decreased turnaround time, and use of a smaller sample volume than manual urine sediment examination. [9][10][11][12] However, these analyzers have low sensitivity and high specificity for cast detection when compared to manual examination of urine sediment 12,13 and have been considered unreliable in some studies. 14 The SediVue Dx ® (SediVue) was the first automated urine analyzer marketed for use in veterinary species. The instrument images approximately 45 high-power fields (HPF/40× microscope objective) of urine sediment, which are then analyzed with image recognition algorithms adapted for dog and cat urine. A study reported that the SediVue analyzer had good agreement with manual sediment examination for most formed elements, including red blood cells (RBC), white blood cells (WBC), and various crystals; however, casts were not evaluated. 8 The objective of this study was to compare the SediVue analyzer with manual sediment examination (i.e., "gold standard" or "reference method") for detection of urinary casts in dog urine samples. We hypothesized that the analyzer would perform similarly to instruments used in human medicine providing low sensitivity and high specificity for cast detection.

| Manual sediment examination
Each urine sample was examined within 30 minutes of collection by a licensed clinical laboratory scientist (CLS). Seven CLS personnel were involved in the study. Well-mixed samples were centrifuged at 360g for 6 minutes (ALC Centrifuge PK 110). Excess supernatant was decanted until 0.2 to 0.5 mL remained to resuspend the pellet. One drop (10 μL) of resuspension was placed on a glass slide and covered with a glass coverslip (22 mm × 22 mm). The entire slide was examined using light microscopy at low power (LPF/10× microscope objective), with attention paid to the coverslip edges for casts. A minimum of 10 highpower fields (HPF/40× microscope objective) were examined by both light and phase contrast microscopy. The CLS then assigned casts to a semiquantitative category per LPF examined: "none"none seen; "rare"not present in every field observed; "few"each field contains a small number (1-2 casts/LPF); "modeach field contains some but not packed (3-5 casts/LPF); "many"each field contains many casts (>5/ LPF). Casts were recorded as hyaline, granular, cellular, waxy, or other.

| SediVue analysis
Urine sediment examination by the SediVue analyzer was performed in tandem with manual microscopy, with 165 μL of well-mixed, uncentrifuged urine manually pipetted into a disposable cartridge by a CLS, while the remaining urine sample was centrifuged for manual microscopy. The CLS was then blinded from the analyzer results. After SediVue centrifugation (30 seconds, 260g), 70 images of the sediment were captured by a camera within the instrument. According to the instrument's manufacturer, these 70 images represent the equivalent of approximately 45 HPF and represent approximately 10 μL of the total 165 μL pipetted into the cartridge. An internal computer system using a convolutional neural network (CNN) (SW 1.0.1.3) algorithm processed the images and identified and quantified any casts present at HPF. The quantification was then converted into number of casts per LPF. Casts were identified as either hyaline or non-hyaline. Nonhyaline casts encompass all casts that are not strictly hyaline (e.g., granular, cellular, waxy, mixed). Cast types were reported as a semiquantitative result: none to rare (NSOR); suspect presence; and >1/ LPF. The SediVue analyzer used for this study was in "research mode," which allowed for the additional reporting of casts per HPF in addition to the semiquantified reporting. This HPF reporting was used to detect samples that had rare casts that would otherwise not be reported by this software version (SW 1.0.1.3) in terms of LPF, as this particular version did not create an LPF extrapolation for "none to rare" and "suspect presence" specimens. All other hardware and software features were identical to that used in clinical practice.
Macroscopic appearance (i.e., color and clarity) and other formed elements found on sediment examination (e.g., WBC, RBC, crystals, epithelial cells, sperm, bacteria, mucus, lipid) were evaluated and recorded for each sample. Macroscopic appearance was subjectively determined by the CLS evaluator. Other formed elements on sediment examination were counted per HPF. Reference intervals for WBC (<3/HPF) and RBC (<5/HPF) were used. Crystals and epithelial cells were recorded per HPF and are described here as "present" if >1/HPF. Bacteria, sperm, and mucus strands were recorded semiquantitatively (none, rare, few, moderate, many) and are described here as "present" if "few," "moderate," or "many" were observed.
Lipid was described as either "present" or "absent."

| SediVue image analysis
Visual analysis of the images captured by the SediVue for all samples considered positive for casts by either methodology (true-positive, false-negative, and false-positive results), and 50 random samples considered negative by both methodologies (true-negative results) was performed by a board-certified veterinary clinical pathologist (SDO) and clinical pathology resident trainee (DMV). A third board-certified veterinary clinical pathologist was consulted if there was disagreement between the 2 reviewers to reach a consensus. the SediVue compared to manual microscopic examination was determined using a threshold of >0/LPF as a positive result (ie, any sample positive for casts). With >0/LPF as a threshold, positive results were those semiquantitatively reported as "rare," "few," "mod," and "many" by manual microscopy, and reported as "NSOR" if any casts were detected per HPF, "suspected presence," and ">1/LPF" by the SediVue.

| Statistical analysis and thresholds
Negative results were those with no casts present on manual microscopy (ie, semiquantitatively reported as "none") and SediVue samples reported as "NSOR" if no casts were detected per HPF. Quantification of casts per HPF was exclusively available in the "research mode" version of the SediVue software as previously described.
Additional sensitivity and specificity values were calculated using a modified threshold of ≥1/LPF for a positive result and <1/LPF for a negative result. Samples semiquantitatively described as having "none" or "rare" casts present by manual methodology, and all samples reported as "NSOR" or "suspect presence" by the SediVue were considered negative results when the modified threshold was used.
Samples semiquantitatively described as "few," "moderate," or "many" casts were considered a positive result by manual methodology, and samples that reported ">1/LPF" were considered a positive result by the SediVue.
The following scale was used to rate sensitivity and specificity,

| Samples
Four hundred fifty-five urine samples were examined of which 12 samples were excluded as they required dilution for SediVue processing, which was not performed. Because the CLS operator was blinded from the analyzer results section where analyzer prompting for dilution is stored, these samples were not diluted and were consequently excluded as they are not acceptable results per the analyzer's instructions. Ultimately, 443 urine samples from 420 dogs were included in the study. The average total volume of submitted urine was 4.6 mL   Table 2 as "normal" or "abnormal" to reflect the typical color (colorless-amber) and clarity (clear-slightly hazy) of urine from normal healthy dogs. 16 There were 108 samples with casts identified by manual microscopy, establishing an overall prevalence of 24.4% (108/443), with a 4.0% (18/443) prevalence of non-hyaline casts in this population.
Three hundred thirty-five samples were negative on manual microscopy. Of the 108 samples positive for casts by manual microscopy, 68.5% (74/108) had "rare" casts present, 23.1% (25/108) had "few" casts present, 7.4% (8/108) had "moderate" numbers of casts present, and 0.9% (1/108) had "many" casts present. Of the 105 samples      (Table 3). Finally, images from 50 random samples of the 288 samples considered negative by both methodologies (true negatives) were also reviewed, and all images reviewed were negative for casts. With these image review findings, a revised sensitivity of 52.0% and revised specificity of 90.6% were determined (Table 1).

| Manual review of SediVue images
Additionally, sensitivity and specificity were calculated based on the reviewed images and using a modified threshold of ≥1/LPF for a positive result. By manual methodology, 34 samples were positive (ie, semiquantitatively described as "few," "moderate," or "many") and 409 samples were negative (ie, "none" or "rare" casts seen

| DISCUSSION
The SediVue is an automated urinalysis sediment analyzer for dog and cat samples. We compared the performance of the SediVue to manual microscopy for detecting urinary casts in fresh dog urine. The ability of the SediVue to detect casts was initially determined to be equivocal, with a low sensitivity (initial Sn = 53.7%) but a moderately high specificity (initial Sp = 86.0%), similar to human automated urine analyzers.
After image review, the sensitivity did not significantly change, but the specificity improved (revised Sp = 90.6%). In samples where casts were in densities commonly considered clinically relevant, ≥1/LPF, the specificity was excellent (modified threshold Sp = 99.3%). Evaluation of the SediVue images revealed areas for analyzer improvement, as well as instances when the analyzer performed better than a human operator.
Overall, the SediVue had moderate agreement with manual methodology (κ = 0.40). In this study, we set an initial cast threshold of >0/LPF as a positive result. However, the SediVue analyzer does not alert the operator that casts are present in the images until "suspected casts" or ">1 cast per LPF" are detected, which is based on a calculation from detections per HPF. We justified the use of >0/LPF because rare casts are of "NSOR." This could potentially lead to delayed diagnosis of systemic or renal diseases, but we recognize that this might also lead to overinterpretation.

Review
In contrast, because 1 to 2 casts/LPF are sometimes observed in normal urine, we also analyzed the sensitivity and specificity when casts were ≥1/LPF (or semiquantitatively described as "few," "moderate," or "many" by manual methodology) to reflect potentially more clinically relevant samples. The sensitivity of the SediVue did not change with the modified threshold (initial revised Sn = 52.0% vs modified threshold Sn = 52.6%), but specificity became excellent (initial revised Sp = 90.6% vs modified threshold Sp = 99.3%). One reason for the improvement in specificity is that the modified threshold classified samples with "rare" casts as true-negative results, which is the majority of samples in the study. Nevertheless, this improvement in specificity demonstrates that the SediVue had few false-positive results claiming casts were present in numbers >1/LPF.
There are a few possible reasons as to why the SediVue analyzer detected more casts than manual microscopy. First, manual microscopy was performed by multiple CLS without confirmation by an additional observer, and consequently, sediment preparation and interpretation could have been subject to error. Confidence in the reference method could have been improved with verification by a second operator or with photomicrographs taken of the evaluated fields.
There was no standard pattern for selection of fields by manual microscopy that was applied across observers. Additionally, the greater number of HPFs examined by the SediVue could have led to increased detections in these cases, although, not enough to improve the overall sensitivity. The idea that the SediVue has a "gentler" centrifugation process leading to increased preservation of formed elements and detection has been proposed, 17 but this seems unlikely, as the g-force used for manual urinalysis (360g) was similar to the g-force used in the SediVue analyzer (260g) in this study. However, the longer centrifugation time for the manual method (6 minutes 18 supporting the idea that the reference method is prone to interobserver imprecision and unreliability as the gold standard.
Because the SediVue and manual sediment exam were performed in tandem, no sample remained to evaluate the precision of the SediVue, and no aliquot of urine was partitioned off beforehand to perform this task.
Interobserver variation was minimized during SediVue image review by using 2 human evaluators and a third evaluator when needed to arrive at a majority decision. Human identification of casts on the images sometimes proved challenging because images were only available at HPF. This truncated the cast of interest at times, creating uncertainty about its identity. Moreover, total length and overall size of the cast was often difficult or impossible to determine because of the image magnification. These challenges could have led to unintentional misidentification of casts by the observers. This is also viewed as a possible limitation for the analyzer, as it might have contributed to the falsepositive results.
This study was performed at a tertiary-care veterinary institution, and case bias might have led to more casts present in urine than anticipated in general veterinary practice. To the author's knowledge, there is no information on the prevalence of casts for a heterogeneous population of dogs, although prevalence of casts in dogs with renal disease has been discussed. 19 The overall prevalence of casts in this Overall, the SediVue had good specificity for cast identification (>0/LPF), but excellent specificity for cast identification when using a more clinically relevant threshold (>1/LPF); however, sensitivity in this study was poor. We postulate that the low sample volume accepted by the SediVue and the low density of casts in the majority of samples could be major contributors to the analyzer's sensitivity. In the future, SediVue images could be compared to photomicrographs taken of the sediment during manual microscopy to improve performance and agreement between the methods. Additionally, improvements to the CNN software to distinguish between cellular debris and mucus could improve the specificity. The SediVue had moderate agreement with manual methodology when performed by a licensed CLS and may provide an option for a rapid initial assessment of the urine sediment, replacing the need for complete sediment examination in some clinical samples. Future studies would also be needed to evaluate the performance of the SediVue compared to veterinary technicians working within a general practice environment.