Abstract
OBJECTIVE To assess sensitivity and specificity of manual and automated measurements of reticulocyte percentage, number, and production index for classification of anemia in dogs.
DESIGN Retrospective case series
SAMPLE 174 blood smears from client-owned dogs with anemia collected between 1993 and 2013 for which reticulocyte parameters were determined manually (nonregenerative anemia, 22; preregenerative anemia, 23; regenerative anemia, 28) or with an automated laser-based counter (nonregenerative anemia, 66; preregenerative anemia, 17; regenerative anemia, 18).
PROCEDURES Diagnostic performance was evaluated with receiver operating characteristic (ROC) curves by considering preregenerative anemia as nonregenerative or regenerative. Sensitivity, specificity, and positive likelihood ratio were calculated by use of cutoffs determined from ROC curves or published reference limits.
RESULTS Considering preregenerative anemia as non regenerative, areas under the curve (AUCs) for reticulocyte percentage, number, and production index were 97%, 93%, and 91% for manual counting and 93%, 90%, and 93% for automated counting. Sensitivity, specificity, and positive likelihood ratio were 82% to 86%, 82% to 87%, and 4.6 to 6.4, respectively. Considering preregenerative anemia as regenerative, AUCs were 77%, 82%, and 80% for manual counting and 81%, 82%, and 92% for automated counting. Sensitivity, specificity, and positive likelihood ratio were 72% to 74%, 76 to 87%, and 2.7 to 6.2, respectively.
CONCLUSIONS AND CLINICAL RELEVANCE Whereas all reticulocyte parameters identified regeneration in anemic dogs, the performance of specific parameters was dependent on the method used. Findings suggested that lower cutoffs than published reference limits are preferred for reticulocyte number and production index and higher cutoffs are preferred for reticulocyte percentage. Reticulocyte production index may be useful when the pretest probability of regeneration is moderate.
Early classification of anemia as regenerative or nonregenerative provides important information on the mechanism of disease and is thus essential for correct diagnosis and management of affected patients.1 When regeneration is detected, acute blood loss or hemolysis should be suspected. Conversely, a lack of regeneration is consistent with several conditions, including hyperacute RBC loss or destruction (ie, preregenerative anemia, with the regenerative response impending), decreased production of erythrocytes as a result of bone marrow disease (eg, hypoplasia or aplasia associated with toxic, infectious, or idiopathic states), or metabolic, neoplastic, inflammatory, and infectious diseases that secondarily affect erythropoiesis.2 The availability of various parameters to detect early regeneration is thus critically important. Because regeneration is defined as the release of immature RBCs into the circulation (except in horses), anemia is usually classified as regenerative or nonregenerative on the basis of detection and quantification of circulating reticulocytes.3
Reticulocyte number is considered a more accurate indicator of regeneration than the percentage of reticulocytes.2,3 Nevertheless, reticulocyte number has several limitations. First, the cutoffs to identify regeneration reported in textbooks and various published studies have been determined by means of manual counts of reticulocytes visualized with supravital stains and are highly variable, with the upper limit of manual counts reported in different studies2–5 varying from 60,000 to 75,000/µL. Second, counts obtained with automated laser-based analyzers are reportedly more sensitive than those obtained manually.6–8 Therefore, the upper limit of reference ranges generated with the automated laser-based analyzers is typically higher than that reported for manual counting, with the upper limit reported in some published studies9–12 as high as 150,000/µL. Thus, the ability of automated reticulocyte counts to differentiate regenerative from nonregenerative anemia varies depending on the method employed for reticulocyte enumeration.13 Third, reticulocyte number alone does not adequately correlate reticulocytosis with severity of anemia, and theoretically, to replace the RBCs lost, the magnitude of reticulocytosis should be inversely related to the severity of anemia.14 To circumvent this limitation, in human medicine, the use of the reticulocyte production index has been proposed, which corrects the percentage of reticulocytes according to the severity of anemia. The reticulocyte production index first corrects the percentage of reticulocytes according to the magnitude of the decrease in Hct, and the corrected reticulocyte percentage is then divided by the time for reticulocyte maturation, expressed in days. This time is inversely proportional to the Hct because, during regeneration, the reticulocytes released into the circulation are immature and remain in circulation for a longer time before becoming mature erythrocytes.15 However, the time required for maturation of circulating reticulocytes in anemic patients has been calculated for humans but not for dogs. Therefore, although we suggest that the rationale for use of the reticulocyte production index is likely to be valid in dogs, the specific calculations may differ. Whereas the time required for reticulocyte maturation in humans and dogs may be similar, this information is currently unknown for dogs. For this reason, a major veterinary hematology textbook15 recommends that the reticulocyte production index not be used “due to lack of information on the impact of various Hct levels on circulating maturation times” because “guidelines for and the clinical usefulness of the corrected reticulocyte percentage or of the reticulocyte production index are not well established in veterinary species.”2 Currently, therefore, evaluation of anisocytosis, polychromasia,3 or reticulocyte number2 has typically been used to estimate the reticulocyte response in affected dogs and cats. Nonetheless, we are not aware of any studies in small animals specifically evaluating the diagnostic accuracy of all reticulocyte parameters. As such, the objective of the study reported here was to assess the diagnostic accuracy of reticulocyte parameters (ie, percentage of reticulocytes, reticulocyte number, and reticulocyte production index) determined by means of either manual counting or an automated laser-based countera in dogs with anemia.
Materials and Methods
Case selection criteria and medical records review
The study was reported according to the STARD 2015 (Standards for Reporting Diagnostic Accuracy Studies) statement.16,17 The STARD 2015 checklist is provided (Supplemental Table S1, available at http://avmajournals.avma.org/doi/suppl/10.2460/javma.249.7.776). Medical records of all dogs with anemia examined at the diagnostic laboratory of the Department of Veterinary Medicine, University of Milan, were retrospectively reviewed. Medical records of anemic dogs, regardless of age, sex, or breed, were included in the study when all of the following were available: CBC and reticulocyte counts or, for records before 2005, good-quality blood smears (eg, smears with well-preserved cell morphology) stained with brilliant cresyl blue for manual reticulocyte count; good-quality Romanowski-stained blood smears (eg, smears with clear nuclear and cytoplasmic details); final diagnosis determined on the basis of history, results of diagnostic testing (eg, serologic testing or PCR assays for infectious or protozoal diseases, cytologic examination, immunophenotyping, Coombs test, or serum biochemical testing), or follow-up information (eg, complete recovery within 1 to 4 weeks after specific treatments or death or failure to improve 1 to 12 months after initial examination); and results of bone marrow cytologic evaluation or stained smears for dogs with nonregenerative anemia.
Cases were excluded on the basis of the following criteria: no CBC or reticulocyte count, no final diagnosis or follow-up information, no stored blood smears (or bone marrow slides in cases of nonregenerative anemia), insufficient data for classification of the type of anemia (eg, Coombs test, anti-RBC antibody positivity, immunophenotyping of leukemia where applicable, results of bone marrow cytologic evaluation, and results of serum biochemical analysis), or blood transfusions prior to blood sample collection. Patients that had received treatments that could potentially interfere with bone marrow regeneration (eg, chemotherapy for lymphoma or leukemia) were also excluded. However, patients treated with agents that do not influence bone marrow activity (eg, treatment for babesiosis or immune-mediated hemolytic anemia) were included.18 Finally, patients with diseases for which anemia can have multiple pathogenic mechanisms (eg, diabetes mellitus) were also excluded.
Hematologic data recorded until 2005 were generated with an impedance counter,a followed by microscopic evaluation of blood smears and manual reticulocyte counts performed on smears stained with brilliant cresyl blue. Data recorded after 2005 were generated by a laser counterb validated in dogs6 that automatically provides reticulocyte counts. Therefore, retrospective analysis of data was performed separately for manual and automated counts. Slides from eligible patients were reviewed by 3 independent observers (2 board-certified clinical pathologists [AG and SP] and a postgraduate student [MM]) to assess whether the morphology of the cells (eg, agglutination, schistocytosis, spherocytosis, leptocytosis, hemoparasites, and leukemic cells) was consistent with the clinical diagnosis in the medical record. This led to the exclusion of samples according to the following criteria: slides with storage artifacts or deterioration, morphological findings inconsistent with the diagnosis (eg, absence of hemoparasites, leukemic cells or signs or regeneration), or insufficient cells to classify the anemia.
Test methods
The index test (the test for which diagnostic performance was being evaluated in the present study) was represented by reticulocyte parameters that were measured (ie, percentage of reticulocytes, reticulocyte number, and reticulocyte production index) in dogs with anemia and determined by means of either manual counting of reticulocytes on brilliant cresyl blue–stained smears or with reticulocyte counts obtained with an automated laser-based counter.a There was no reference test for categorizing anemia in dogs as regenerative or nonegenerative. Therefore, patient history, initial clinical signs, and results of repeated CBCs and other relevant laboratory tests suggestive of conditions typically associated with regenerative anemia (eg, hemorrhage and hemoparasites), together with follow-up information, including response to treatment (ie, recovery within 1 to 4 weeks vs failure to improve and death or euthanasia), were used as the reference standard.
Calculation of reticulocyte parameters
For both manual and automated counts, patients were classified into 3 groups (regenerative, nonregenerative, and preregenerative anemia). Hematocrit (%), number of erythrocytes (X 106/µL), and percentage of reticulocytes were used to calculate reticulocyte number and reticulocyte production index as follows:
The reticulocyte production index was calculated according to the formula used in humans15 by dividing the corrected reticulocyte percentage by the reticulocyte maturation time. For dogs, we estimated that the reticulocyte maturation time increased by 0.5 days for each 10-point decrease in Hct, assuming that for an Hct of 45% (considered within reference limits in dogs), the maturation time was 1 day. Therefore, the reticulocyte production index was calculated only for cases in which the Hct was < 45%.
Statistical analysis
Data on percentage of reticulocytes, reticulocyte numbers, and reticulocyte production index for each group (ie, regenerative, nonregenerative, and preregenerative) and counting method (ie, manual and automated) were analyzed with the Friedman ANOVA test, followed by the Bonferroni method to estimate 95% confidence intervals. The diagnostic accuracy of reticulocyte parameters was assessed by dividing patients into regenerative and nonregenerative anemia groups via 2 approaches. First, dogs with regenerative anemia comprised the regenerative anemia group, and dogs with nonregenerative or preregenerative anemia comprised the nonregenerative anemia group, which assumed that dogs with preregenerative anemia did not have strong regeneration at the moment of sampling, even if a stimulus toward bone marrow regeneration had already been initiated. Next, dogs with nonregenerative anemia were included in the nonregenerative anemia group, and dogs with regenerative and preregenerative anemia were included in the regenerative anemia group. Thus, for this analysis, it was assumed that dogs with preregenerative anemia had circulating reticulocytes, albeit presumably in low numbers. The rationale for this approach was to assess the ability of reticulocyte parameters to detect early regeneration.
To assess diagnostic accuracy, for each of these approaches and for each reticulocyte parameter, numbers of true-positive, false-positive, true-negative, and false-negative results were calculated. True-positive samples were those from the regenerative anemia group with reticulocyte values higher than each operating point, whereas true-negative samples were those from the nonregenerative anemia group with reticulocyte values lower than each operating point. The term operating point refers to each numerical value recorded in the study.19 False-positive samples were those from the nonregenerative anemia group with reticulocyte values higher than each operating point, with false-negative samples being those from the regenerative anemia group with reticulocyte values lower than each operating point. Sensitivity and specificity were calculated by means of standard formulae,20 and the positive likelihood ratio was obtained with the following formula:
Finally, ROC curves were obtained by plotting sensitivity versus 1 – specificity to determine the ability of reticulocyte parameters to identify dogs in the regenerative anemia group.19 The AUCs of the ROC curves for each parameter were compared. Moreover, for each parameter and for each counting method (ie, manual or automated), the optimal cutoff corresponding to the value closest to the upper left corner of the graph was identified. Sensitivity, specificity, and positive likelihood ratio were calculated by use of the cutoffs determined with the ROC curves and then with the upper reference limits from our laboratory as cutoffs, established through the robust method on the basis of results for healthy nonanemic dogs (for manual counts, n = 130; for automated counts, 146) and as recommended by American Society for Veterinary Clinical Pathology guidelines for establishing reference intervals.22 For reticulocyte production index, a cutoff of 1.0 was used, which was the cutoff commonly used in humans.15 Values of P ≤ 0.05 were considered significant. All statistical analyses were performed with commercial software.c
Results
Case selection and characteristics of groups
Flow of participants through the study is illustrated (Figure 1). A total of 174 cases (manual counting, n = 73; automated counting, 101) were eligible for inclusion. The group of dogs with nonregenerative anemia (n = 88; 22 dogs for manual counting and 66 for automated counting) included dogs of 32 breeds (43 sexually intact males, 29 sexually intact females, 12 spayed females, and 4 neutered males; median age, 8 years; range, 2 months to 16 years [manual counting: median age, 3.5 years; range 2 months to 13 years; automated counting: median age, 9 years; range, 8 months to 16 years). Dogs included in this group had hematologic neoplasms or other tumors (n = 37), inflammation (25), iron deficiency anemia (6), chronic kidney disease (14), hepatic failure (4), estrogen toxicosis (1), or aplastic anemia (1). All of these dogs were anemic with values lower than the laboratory reference ranges (RBC number, 5.5 × 1012/L to 8.8 × 1012/L; Hct, 0.37 to 0.51 L/L; hemoglobin concentration, 120 to 180 g/L) and did not have positive results for anisocytosis or polychromasia on smears. In 21 of 88 cases, results of bone marrow cytologic evaluation confirmed erythroid hypoplasia. Despite improvement of clinical signs following treatment of the primary disease, erythrocyte counts remained below reference limits for an extended period for all dogs in this group (ie, 1 week to 12 months, depending on the patient and specific disease).
Study flow diagram illustrating the selection of cases in a study assessing the sensitivity and specificity of manual and automated measurement of reticulocyte percentage, number, and production index for classification of anemia in dogs.
Citation: Journal of the American Veterinary Medical Association 249, 7; 10.2460/javma.249.7.776
The group of dogs with regenerative anemia (n = 46; 28 dogs for manual counting and 18 for automated counting) included dogs of 19 breeds (21 sexually intact males, 14 sexually intact females, 6 spayed females, and 5 neutered males; median age, 6.5 years, range, 8 months to 15 years [manual counting: median age, 5.5 years, range 8 months to 15 years; automated counting: median age: 7 years, range 1 to 12 years]). This group included dogs that had diseases that caused blood loss or hemolysis, with clinical signs present for > 4 to 5 days. Dogs were anemic, had evidence of anisocytosis and polychromasia on blood smears, and had normalization of erythroid parameters (RBC counts, Hct, and hemoglobin concentration) within 1 to 4 weeks after treatment. This time span was variable because normalization was rapid (ie, approx 1 week) in dogs with mild anemia, but slower in dogs with severe anemia as a result of the primary disease. The most common condition in this group was immune-mediated hemolytic anemia (n = 27) characterized by macro- or microagglutination, spherocytes, schistocytes, anti-RBC antibodies, or a combination of these findings. In 17 of these 27 dogs (10 with manual counts and 7 with automated counts), immune-mediated hemolytic anemia was associated with other primary diagnoses, typically of inflammatory origin, whereas in the remaining 10 dogs, the hemolytic anemia was considered autoimmune on the basis of a positive response to immunosuppressive drugs. The regenerative anemia group also included 6 dogs with babesiosis diagnosed by means of detection of Babesia spp in RBCs and 13 dogs with anemia secondary to blood loss that occurred at least 5 days prior to collection of the blood sample. In patients in this group, hemorrhage or the hemolytic crises resolved after appropriate treatment and erythrocyte mass was restored in a few days.
The group of dogs with preregenerative anemia (n = 40; 23 dogs for manual counting and 17 for automated counting) included dogs of 19 breeds (13 sexually intact males, 15 sexually intact females, 6 spayed females, and 6 neutered males [manual counting: median age, 5 years; range, 1 to 12 years; automated counting: median age, 6 years; range, 3 months to 14 years]). Dogs included in this group had severe postsurgical or post-traumatic hemorrhage (ie, dogs with visible evidence of acute hemorrhage; n = 10) or clinical signs consistent with hemolytic crises (eg, fever, hemoglobinuria, icterus, splenomegaly; 30), with blood samples obtained no more than 2 days after the onset of clinical signs. In most cases (23/30), hemolysis was associated with a diagnosis of babesiosis and resolved after appropriate treatment, whereas in other cases (7/30), no cause for the hemolysis was found and an autoimmune condition was suspected on the basis of response to immunosuppressive treatments. Dogs were included in this group regardless of the presence of anemia, anisocytosis, or polychromasia. In 5 dogs, evidence of anemia was absent at the time of the first blood sample collection, but was detected when a second sample was collected 1 day later; reticulocytosis and restoration of erythrocyte mass occurred within a week after initiation of appropriate treatment. In these patients, test results for the first blood sample (ie, during the preregenerative phase) were included in the study.
Reticulocyte parameters
Distributions of reticulocyte parameters in the 3 groups after manual and automated counting were determined (Figure 2). After manual counting, all reticulocyte parameters were significantly (P < 0.001) higher for dogs with regenerative anemia, compared with values for dogs with nonregenerative or preregenerative anemia. Moreover, in all 28 dogs with regenerative anemia for which manual counting was performed, the percentage of reticulocytes was > 0.6% (ie, the upper reference limit for the laboratory), and in 27 of these 28 dogs, the number of reticulocytes was higher than the upper reference limit (46.0 × 109/L). Proportions of dogs with a reticulocyte percentage or reticulocyte number higher than the upper reference limit were significantly lower for the preregenerative anemia group (11/23 and 7/23 dogs, respectively) and nonregenerative anemia group (8/22 and 3/22 dogs, respectively) than for the regenerative anemia group (28/28 and 27/28, respectively).
Distribution of reticulocyte percentage, reticulocyte number (expressed as × 109/L), and reticulocyte production index determined by means of manual [A] or automated [B] counting methods in dogs (n = 174) with nonregenerative, preregenerative, or regenerative anemia. Boxes indicate the interquartile range (IQR), the horizontal line indicates the median, and the whiskers extend to the first quartile – (1.5 × IQR) and the third quartile + (1.5 × IQR). Open circles indicate near outliers (values exceeding the third quartile + [1.5 × IQR]), and solid circles indicate far outliers (values exceeding the third quartile + [3.0 × IQR]). The horizontal dashed gray line represents the reference interval for the laboratory. Symbols within boxes indicate significant differences, compared with nonregenerative anemia (*P < 0.05; **P < 0.01; ***P < 0.001) or with preregenerative anemia (††P < 0.01; †††P < 0.001).
Citation: Journal of the American Veterinary Medical Association 249, 7; 10.2460/javma.249.7.776
After automated counting, all reticulocyte parameters were significantly higher for dogs with regenerative or preregenerative anemia, compared with values for dogs with nonregenerative anemia (for regenerative anemia vs nonregenerative anemia groups, P < 0.001 for all reticulocyte parameters; for preregenerative anemia vs nonregenerative anemia groups, P < 0.001 for reticulocyte production index, P < 0.05 for reticulocyte number, and P < 0.05 for reticulocyte percentage). The reticulocyte percentage was significantly (P < 0.05) higher in dogs with regenerative anemia than in dogs with preregenerative anemia. All 18 dogs with regenerative anemia had a reticulocyte percentage that was higher than the upper reference limit, whereas reticulocyte numbers were within the reference interval in 2 of these 18 dogs. In contrast, reticulocyte numbers were within the reference interval in 43 of the 66 dogs with nonregenerative and in 4 of the 17 dogs with preregenerative anemia. In addition, 46 of the 66 dogs with nonregenerative anemia and 14 of the 17 dogs with preregenerative anemia had a reticulocyte percentage that was higher than the upper reference limit.
Sensitivity and specificity of reticulocyte parameters
Sensitivity, specificity, and positive likelihood ratio for manual and automated counting were summarized (Table 1). The AUCs when dogs with preregenerative anemia were included in the nonregenerative anemia group versus when they were included in the regenerative anemia group for manual or automated counting were calculated (Figure 3). In all instances, the AUCs of the reticulocyte parameters were significantly (P < 0.05) higher than the line of no discrimination.
Comparison of ROC curves for reticulocyte percentage (squares), reticulocyte number (diamonds), and reticulocyte production index (circles). Measurements were obtained with manual (A and B) or automated (C and D) counting methods. Dogs with preregenerative anemia were included in the nonregenerative anemia group (A and C) or the regenerative anemia group (B and D).
Citation: Journal of the American Veterinary Medical Association 249, 7; 10.2460/javma.249.7.776
Sensitivity, specificity, and positive likelihood ratio for reticulocyte percentage, reticulocyte number, and reticulocyte production index, with cutoff values determined by analysis of ROC curves for dogs (n = 174) with anemia examined between 1993 and 2013.
| Manual counting (n = 73) | Automated counting (n = 101) | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Variable | AUC | Cutoff | Sensitivity (%) | Specificity (%) | Positive ikelihood ratio | AUC | Cutoff | Sensitivity (%) | Specificity (%) | Positive likelihood ratio | |
| Preregenerative anemia included with nonregenerative anemia | Reticulocyte percentage | 0.97 (0.94–1.00) | 1.70 | 85.7 (67.3–96.0) | 86.7 (73.2–94.9) | 6.40 | 0.93 (0.87–0.98) | 2.28 | 82.4 (56.6–96.2) | 83.3 (73.6–90.6) | 4.94 |
| Reticulocyte No. | 0.93 (0.88–0.98)* | 57.40 | 82.1 (63.1–93.9) | 82.2 (67.9–92.0) | 4.62 | 0.90 (0.82–0.98) | 85.90 | 82.4 (56.6–96.2) | 82.1 (72.3–89.6) | 4.61 | |
| RPI | 0.91 (0.85–0.98)* | 0.58 | 82.1 (63.1–93.3) | 82.9 (67.9–92.8) | 4.81 | 0.93 (0.88–0.98) | 0.41 | 82.4 (56.6–96.2) | 83.8 (73.8–91.1) | 5.07 | |
| Preregenerative anemia included with regenerative anemia | Reticulocyte | 0.77 (0.66–0.88) | 0.75 | 72.5 (58.3–84. l) | 72.7 (49.8–89.3) | 2.66 | 0.81 (0.72–0.91) | 1.42 | 73.5 (55.6–87.1) | 76.1 (64.1–85.7) | 3.08 |
| percentage Reticulocyte No. | 0.82 (0.72–0.92)† | 35.82 | 72.5 (58.3–84. l) | 72.7 (49.8–89.3) | 2.66 | 0.82 (0.73–0.91) | 63.78 | 73.5 (55.6–87.1) | 77.6 (65.8–86.9) | 3.28 | |
| RPI | 0.80 (0.70–0.91) | 0.33 | 74.5 (59.7–86. l) | 72.7 (49.8–89.3) | 2.73 | 0.92 (0.86–0.97)‡ | 0.29 | 83.3 (65.3–94.4) | 86.6 (76.0–93.7) | 6.20 | |
Values in parentheses are 95% confidence intervals.
Significantly (P < 0.05) different from value for reticulocyte percentage.
Significantly (P < 0.01) different from value for reticulocyte percentage.
Significantly (P < 0.01) different from values for reticulocyte number and reticulocyte percentage.
RPI = Reticulocyte production index.
For both manual and automated counting, the discriminating power of all reticulocyte parameters, as defined by the AUCs, was higher when dogs with preregenerative anemia were included in the nonregenerative anemia group than when dogs with preregenerative anemia were included in the regenerative anemia group. When dogs with preregenerative anemia were included in the nonregenerative anemia group, the AUC for the percentage of reticulocytes was significantly (P ≤ 0.05) higher than that for both reticulocyte number and reticulocyte production index for manual counting, but no significant differences among the 3 AUCs were found after automated counting. In contrast, when dogs with preregenerative anemia were included in the regenerative anemia group, the highest AUC was recorded for reticulocyte number with manual counting (P < 0.01 vs reticulocyte percentage) and for reticulocyte production index with automated counting (P < 0.01 vs reticulocyte number and reticulocyte percentage).
At the cutoff determined by the ROC curves, the sensitivity and specificity of reticulocyte percentage, reticulocyte number, and reticulocyte production index ranged from 82% to 87% when dogs with preregenerative anemia were included in the nonregenerative anemia group, and from 72% to 77% when dogs with preregenerative anemia were included in the regenerative anemia group, except that reticulocyte production index had sensitivity and specificity > 80% with automated counting. For all parameters, the positive likelihood ratio was > 1, indicating that in the case of values higher than the cutoffs, the probability of having regeneration was 4.6 to 6.4 times the probability of not having regeneration when dogs with preregenerative anemia were included in the nonregenerative anemia group. The probability of having regeneration was 2.7 to 6.2 times the probability of not having regeneration when dogs with preregenerative anemia were included in the regenerative anemia group.
Diagnostic performance of reticulocyte percentage and reticulocyte number at the cutoff corresponding with the upper reference limit of the laboratory and of reticulocyte production index at the cutoff of 1.0, which is the published recommendation for reticulocyte production index in humans,15 was summarized (Table 2). When these cutoff values were used, instead of the cutoffs established with ROC curves, and dogs with preregenerative anemia were included in the nonregenerative anemia group, sensitivity of reticulocyte percentage increased and specificity and positive likelihood ratio decreased, independent of counting method (manual or automated). Sensitivity and positive likelihood ratio of reticulocyte number increased and specificity decreased with manual counting, and sensitivity of reticulocyte number decreased and specificity and positive likelihood ratio increased with automated counting. Lastly, sensitivity of reticulocyte production index decreased and specificity and positive likelihood ratio increased, independent of counting method. When dogs with preregenerative anemia were included in the regenerative anemia group, specificity of reticulocyte percentage increased and sensitivity and positive likelihood ratio decreased with manual counting. However, for the other reticulocyte parameters determined with manual counting and for all reticulocyte parameters determined with automated counting, sensitivity decreased and specificity and positive likelihood ratio increased.
Sensitivity (Sens [%]), specificity (Spec [%]), and positive likelihood ratio (LR+) of the percentage and number of reticulocytes (expressed as x109/L) at the cutoff corresponding to the upper limit of the reference interval of the laboratory as well as the reticulocyte production index at a cutoff of 1.0 reported in the literature for humans.15
| Manual counting (n = 73) | Automated counting (n = 101) | |||||||
|---|---|---|---|---|---|---|---|---|
| Variable | Cutoff | Sens | Spec | LR+ | Cutoff | Sens | Spec | LR+ |
| Preregenerative anemia included with nonregenerative anemia | ||||||||
| Reticulocyte percentage > 0.6% | 100.0 | 55.6 | 2.2 | Reticulocyte percentage > 1.8% | 88.2 | 77.4 | 3.9 | |
| Reticulocyte number > 46.0 × 100/L | 96.4 | 80.0 | 4.8 | Reticulocyte number > 120.5 × 100/L | 58.8 | 92.9 | 8.2 | |
| Reticulocyte production index > 1.0 | 64.3 | 90.2 | 6.6 | Reticulocyte production index > 1.0 | 52.9 | 97.5 | 21.1 | |
| Preregenerative anemia included with regenerative anemia | ||||||||
| Reticulocyte percentage > 0.6% | 76.5 | 63.6 | 2.1 | Reticulocyte percentage > 1.8% | 61.8 | 80.0 | 3.2 | |
| Reticulocyte number > 46.0 × 100/L | 66.7 | 86.4 | 4.9 | Reticulocyte number > 120.5 × 100/L | 38.2 | 95.5 | 8.5 | |
| Reticulocyte production index > 1.0 | 44.7 | 95.5 | 9.8 | Reticulocyte production index > 1.0 | 44.4 | 97.6 | 18.4 | |
Data were generated on the basis of hematologic test results for blood samples from 174 dogs with nonregenerative (n = 88), preregenerative (40), or regenerative (46) anemia examined between 1993 and 2013 for which measurements were obtained manually or with an automated laser-based cell counter.a
Discussion
Results of the present study indicated that reticulocyte percentage, reticulocyte number, and reticulocyte production index could all be used to identify dogs with regenerative versus nonregenerative anemia, but performance of these parameters depended on the counting method used (ie, manual vs automated) and the intensity of regeneration (ie, whether preregenerative anemia was considered regenerative or nonregenerative). Therefore, the test of choice and the information provided by the various tests varied with the clinical scenario (Appendix).
Although some laser-based hematology analyzers provide innovative reticulocyte parameters to detect early regeneration,23 in routine clinical practice, the diagnosis of regenerative anemia is usually made on the basis of quantification of reticulocyte responses.1–3 Reticulocytes may be expressed as either a percentage or absolute number. The reticulocyte production index corrects for the magnitude of reticulocytosis according to the severity of anemia, because regeneration should be more intense in patients with severe anemia.1,2,14 Therefore, the reticulocyte production index should be the best marker of regeneration. However, the reticulocyte production index is calculated on the basis of the maturation time of human reticulocytes, which may not accurately reflect regeneration in dogs. Accordingly, reticulocyte number is usually preferred in dogs.1,2,15 However, no studies have evaluated the diagnostic accuracy of these parameters. The present investigation was designed to assess the diagnostic accuracy of reticulocyte percentage, number, and production index to detect regeneration in dogs. Data for corrected reticulocyte percentage (ie, the percentage of reticulocytes corrected for the degree of anemia) were not presented because veterinarians are generally not familiar with this parameter and no reference intervals are available, thus hampering our ability to evaluate its diagnostic accuracy in dogs.
Unfortunately, there is no gold-standard test for categorizing anemia in dogs as regenerative or nonregenerative in clinical practice. Detection of erythroid hyperplasia by means of bone marrow cytologic evaluation closely resembles a gold standard, but erythroid hyperplasia may also be seen with iron deficiency anemia,24 pure red cell aplasia,25 and myelodysplastic syndromes;26 moreover, bone marrow cytologic evaluation is not recommended in routine practice once the cause of blood loss or RBC destruction has been identified.2 Therefore, this method cannot be applied in a clinical study, except when nonregenerative anemia is suspected. Theoretically, other statistical tests may be applied to evaluate this;27 however, the number of cases in the present study was too small to consider such methods. Therefore, in this study, cases were classified as regenerative or nonregenerative on the basis of clinical findings (eg, presence of conditions associated with regenerative anemia such as hemolysis or hemorrhage) and follow-up information, including response to treatment and restoration of erythroid values within 1 to 4 weeks, which indicated an efficient erythroid response. Moreover, strict exclusion criteria were applied to exclude dogs with diseases that depend on multiple or unclear pathogenic mechanisms and atypical cases associated with regenerative or nonregenerative anemia. Because of this, many cases from the medical records database were excluded, and only samples with precise information on the presence or absence of regeneration or with a diagnosis of acute blood loss documented before regeneration became fully evident (ie, preregenerative anemia) were included. Preregenerative anemia was diagnosed if anemia was present at the time of blood sample collection or on the following day and if reticulocytosis and restoration of erythrocyte mass occurred within a week.
Considering these aspects, we suggest that the study population was adequate to investigate the diagnostic accuracy of reticulocyte parameters because it included patients with diseases characterized by full regeneration, inappropriate reticulocytosis (ie, iron deficiency), or no regeneration. The number of cases of regenerative anemia submitted to our institution has decreased in recent years, when reticulocytes were routinely counted with a laser-based cell counter. As a result, the diagnostic performance of reticulocyte parameters was expressed in terms of sensitivity, specificity, and positive likelihood ratio, but not in terms of predictive values, which are affected by the prevalence of the condition under investigation.21
The highest reticulocyte numbers and percentages of reticulocytes were found, as expected, in the group of dogs with regenerative anemia. In addition, the group of dogs with preregenerative anemia had values that were higher than values determined with automated counting for the group of dogs with nonregenerative anemia, which we suggest is likely because of the high analytic sensitivity of the laser-based analyzer.7 Reference intervals for percentage of reticulocytes and reticulocyte number were higher for automated counting, as previously reported.9–12 Therefore, the proportion of dogs with values higher than reference intervals varied with the method; only the percentage of reticulocytes in dogs with regenerative anemia was consistently higher than the reference interval, whereas reticulocyte numbers were within reference limits in some dogs with regenerative anemia (1/28 with manual counting and 2/18 with automated counting). This suggested that comparison of reticulocyte numbers alone with reference intervals may be misleading.
In the present study, we first calculated the sensitivity and specificity by including dogs with preregenerative anemia in the nonregenerative anemia group to assess how each parameter identifies an intense regeneration. We then included dogs with preregenerative anemia in the regenerative anemia group to simulate the condition in which RBC destruction or loss is clinically evident and early regeneration must be investigated (a scenario that requires higher diagnostic accuracy of reticulocyte counts). All parameters discriminated dogs with regenerative anemia, especially if preregenerative anemia was included in the nonregenerative anemia group, suggesting that, as expected, the utility of reticulocyte parameters is higher when regeneration is fully active. The ROC curves suggested that manual counting of the percentage of reticulocytes better identified full regeneration than reticulocyte number during early regeneration, whereas with automated counting, the reticulocyte production index best identified both intense and early regeneration.
As expected, the optimal cutoffs defined by the ROC curves in the present study were different for manual and automated counting and were higher when dogs with preregenerative anemia were included in the nonregenerative anemia group. In the latter scenario, the cutoff for the percentage of reticulocytes was higher than the upper reference limit of our laboratory, whereas the cutoff for the number of reticulocytes was similar (manual) or lower (automated) than the upper reference limit. When dogs with preregenerative anemia were included in the regenerative anemia group, the cutoffs for the percentage of reticulocytes were close to or lower than the reticulocyte production index. Notably, the optimal cutoff for reticulocyte production index was lower than the published recommended cutoff of 1.0 for humans.15
In the present study, the positive likelihood ratio was not particularly high; however, on the basis of the Bayesian approach,21 all parameters may be clinically useful when the pretest probability of regeneration is high (eg, external blood loss and clinical evidence of hemolysis). However, the cutoffs defined by ROC curves maximize both sensitivity and specificity.21 In routine clinical practice, it may be preferable to use a test with greater sensitivity or specificity, depending on the clinical signs. If the pretest probability of regeneration is high, we suggest that it would be preferable to use a test that avoids false-positive results and that can provide a confirmatory test. From this perspective, it is worth noting that for both clinical scenarios included in this study (ie, patients with preregenerative anemia included in the nonregenerative anemia group vs patients with preregenerative anemia included in the regenerative anemia group) and independent of the method used for reticulocyte counting, the reticulocyte production index when compared with the published cutoff value published for humans15 and, to a lesser extent, the reticulocyte number when compared with the upper reference limit for our laboratory had high specificity. Moreover, tests with a high likelihood ratio are particularly useful when the pretest probability is intermediate (eg, 50%).21 An intermediate situation may be present when no history is available and when signs of blood loss or hemolysis are unclear. From this perspective, the reticulocyte production index may be the most informative parameter. In dogs with a reticulocyte production index > 1.0, the probability of having regeneration is much higher (up to 21 times with automated counting), compared with the probability of not having regeneration.
Limitations of the present study included the lack of a true reference method and the relatively low caseload, compared with the initial number of cases included in the database. As stated, bone marrow analysis, which may be considered the ideal reference method, is currently not applicable in clinical practice. Therefore, we classified cases as regenerative or not on the basis of follow-up information, which ultimately may be considered the best alternative method to assess whether erythroid mass is restored over time. Regarding the low caseload, the application of strict inclusion criteria led to a reduction of the caseload from approximately 7,000 to < 200 cases. This allowed us to not overinterpret the results of the study. However, it may be advisable in the future to standardize the diagnostic approach in anemic dogs and to increase the caseload in studies such as this one.
In the present study, whereas all reticulocyte parameters identified regeneration in anemic dogs, the performance of the various parameters was dependent on the counting method used and on the intensity of regeneration. In general, compared with published reference limits, we suggest that lower cutoffs may be preferred for reticulocyte number and reticulocyte production index, and higher cutoffs may be preferred for reticulocyte percentage. By use of these cutoffs, defined by ROC curves, the percentage of reticulocytes and, to a lesser extent, reticulocyte number will identify full or early regeneration with manual counting, whereas the reticulocyte production index maximizes the possibility to detect full and early regeneration with automated counting. From a clinical perspective, however, at these cutoffs, all parameters will serve as confirmatory tests only when the pretest probability of regeneration is high. When the probability of regeneration cannot be established, a reticulocyte production index > 1.0, independent of the counting method, may be the best marker of regeneration in dogs with anemia.
Acknowledgments
Supported in part by the Dote Ricerca grant through the European Social Fund (Fondo Sociale Europeo, Regione Lombardia).
Presented in part at the 13th Conference of the European Society of Veterinary Clinical Pathology and the European College of Veterinary Clinical Pathology, Dublin, Ireland, August 2011.
The authors declare that there were no conflicts of interest.
ABBREVIATIONS
| AUC | Area under the curve |
| CI | Confidence interval |
| ROC | Receiver operating characteristic |
Footnotes
Sysmex XT 2000-iV, Sysmex Corporation, Kobe, Japan.
Hemat 8, SEAC, Calenzano, Firenze, Italy.
Analyse-it, version 2.30, Analyse-it Software Ltd, Leeds, England.
References
1. Cowgill ES, Neel JA, Grindem CB. Clinical application of reticulocyte counts in dogs and cats. Vet Clin North Am Small Anim Pract 2003; 33: 1223–1244.
2. Tvedten H. Laboratory and clinical diagnosis of anemia. In: Weiss DJ, Wardrop KJ, eds. Schalm's veterinary hematology. 6th ed. Ames, Iowa: Wiley-Blackwell, 2010;152–161.
3. Hodges J, Christopher MM. Diagnostic accuracy of using erythrocyte indices and polychromasia to identify regenerative anemia in dogs. J Am Vet Med Assoc 2011; 238: 1452–1458.
4. Weiss DJ. Bone marrow pathology in dogs and cats with non-regenerative immune-mediated haemolytic anaemia and pure red cell aplasia. J Comp Pathol 2008; 138: 46–53.
5. Rizzi TE, Meinkhot JH, Clinkendeard KD. Normal hematology of the dog. In: Weiss DJ, Wardrop KJ, eds. Schalm's veterinary hematology. 6th ed. Ames, Iowa: Wiley-Blackwell, 2010;799–810.
6. Lilliehöök I, Tvedten H. Validation of the Sysmex XT-2000iV hematology system for dogs, cats, and horses. I. Erythrocytes, platelets, and total leukocyte counts. Vet Clin Pathol 2009; 38: 163–174.
7. Moritz A, Becker M. Automated hematology systems. In: Weiss DJ, Wardrop KJ, eds. Schalm's veterinary hematology. 6th ed. Ames, Iowa: Wiley-Blackwell, 2010;1054–1066.
8. Tvedten H, Moritz A. Reticulocyte and Heinz body staining and enumeration. In: Weiss DJ, Wardrop KJ, eds. Schalm's veterinary hematology. 6th ed. Ames, Iowa: Wiley-Blackwell, 2010;1067–1073.
9. Moritz A, Fickenscher Y, Meyer K, et al. Canine and feline hematology reference values for the ADVIA 120 hematology system. Vet Clin Pathol 2004; 33: 32–38.
10. Bourgès-Abella N, Geffré A, Concordet D, et al. Canine reference intervals for the Sysmex XT-2000iV hematology analyzer. Vet Clin Pathol 2011; 40: 303–315.
11. Campora C, Freeman KP, Serra M, et al. Reference intervals for Greyhounds and lurchers using the Sysmex XT-2000iV hematology analyzer. Vet Clin Pathol 2011; 40: 467–474.
12. Serra M, Freeman KP, Campora C, et al. Establishment of canine hematology reference intervals for the Sysmex XT-2000iV hematology analyzer using a blood donor database. Vet Clin Pathol 2012; 41: 207–215.
13. Collicutt NB, Grindem CB, Neel JA. Comparison of manual polychromatophilic cell and automated reticulocyte quantification in evaluating regenerative response in anemic dogs. Vet Clin Pathol 2012; 41: 256–260.
14. Riley RS, Ben-Ezra JM, Goel R, et al. Reticulocytes and reticulocyte enumeration. J Clin Lab Anal 2001; 15: 267–294.
15. Christian JA. Erythrokinetics and erythrocyte destruction. In: Weiss DJ, Wardrop KJ, eds. Schalm's veterinary hematology. 6th ed. Ames, Iowa: Wiley-Blackwell, 2010;136–143.
16. Bossuyt PM, Reitsma JB, Bruns DE, et al. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ 2015; 351: h5527.
17. Bossuyt PM, Reitsma JB, Bruns DE, et al. The STARD statement for reporting studies of diagnostic accuracy: explanation and elaboration. Clin Chem 2003; 49: 7–18.
18. Weiss DJ. Drug-induced blood cell disorders. In: Weiss DJ, Wardrop KJ, eds. Schalm's veterinary hematology. 6th ed. Ames, Iowa: Wiley-Blackwell, 2010;98–105.
19. Gardner IA, Greiner M. Receiver-operating characteristic curves and likelihood ratios: improvements over traditional methods for the evaluation and application of veterinary clinical pathology tests. Vet Clin Pathol 2006; 35: 8–17.
20. Christenson RH. Evidence-based laboratory medicine—a guide for critical evaluation of in vitro laboratory testing. Ann Clin Biochem 2007; 44: 111–130.
21. Albert A. On the use and computation of likelihood ratios in clinical chemistry. Clin Chem 1982; 28: 1113–1119.
22. Friedrichs KR, Harr KE, Freeman KP, et al. ASVCP reference interval guidelines: determination of de novo reference intervals in veterinary species and other related topics. Vet Clin Pathol 2012; 41: 441–453.
23. Fry MM, Kirk CA. Reticulocyte indices in a canine model of nutritional iron deficiency. Vet Clin Pathol 2006; 35: 172–181.
24. Weiss DJ. Iron and copper deficiencies and disorders of iron metabolism. In: Weiss DJ, Wardrop KJ, eds. Schalm's veterinary hematology. 6th ed. Ames, Iowa: Wiley-Blackwell, 2010;167–171.
25. Weiss DJ. Pure red cell aplasia. In: Weiss DJ, Wardrop KJ, eds. Schalm's veterinary hematology. 6th ed. Ames, Iowa: Wiley-Blackwell, 2010;251–255.
26. Weiss DJ. Myelodysplastic syndromes. In: Weiss DJ, Wardrop KJ, eds. Schalm's veterinary hematology. 6th ed. Ames, Iowa: Wiley-Blackwell, 2010;467–474.
27. Johnson WO, Gastwirth JL, Pearson LM. Screening without a “gold standard”: the Hui Walter model revisited. Am J Epidemiol 2001; 153: 921–924.
Appendix
Summary of recommendations for interpretation of reticulocyte parameters in dogs with anemia when the pretest probability of regeneration is high (eg, blood loss or hemolysis occurring at least 4 days prior to blood sample collection) or moderate (recent or hyperacute blood loss or hemolysis).
| Pretest probability | Counting method | Cutoffs established by ROC curves | Cutoffs established according to laboratory reference limits or published literature15 |
|---|---|---|---|
| High | Manual | A reticulocyte percentage > 1.7% strongly suggests active regeneration (LR = 6.4) | A reticulocyte percentage < 0.6% or reticulocyte number < 46 × 109/L excludes active regeneration (sensitivity, 100% and 96%, respectively) |
| A reticulocyte production index > 1.0 suggests active regeneration (LR = 6.6) | |||
| Automated | A reticulocyte production index > 0.41 strongly suggests active regeneration (LR > 5.1) | A reticulocyte production index > 1.0 strongly suggests active regeneration (LR > 21.1) | |
| Moderate | Manual | A reticulocyte production index > 0.33 may suggest early regeneration (LR > 2.7) | A reticulocyte production index > 1.0 strongly suggests early regeneration (LR > 9.8) |
| Automated | A reticulocyte production index > 0.29 suggests early regeneration (LR > 6.2) | A reticulocyte production index > 0.29 suggests early regeneration (LR > 18.4) |
LR = Positive likelihood ratio.
