Diagnosis of EPM caused by Sarcocystis neurona traditionally has been based on sequential immunoblot testing of serum and CSF for antibodies against the parasite.1 Horses with clinical signs of neurologic disease and seropositive for anti–S neurona antibodies are considered infected with the parasite in the CNS and a diagnosis of EPM is made if a positive test result for antibodies in CSF is obtained.1 Testing of CSF to confirm S neurona infection has been justified by the high prevalence of healthy seropositive horses and, therefore, low specificity and positive predictive value of the immunoblot serologic test.2–6 Anti–S neurona antibodies in uncontaminated (ie, minimal RBCs) CSF samples from horses without evidence of blood-brainbarrier damage (albumin quotient in the reference range) and a normal to high ratio of IgG in serum and CSF (IgG index) are considered indicative of intrathecal production of antibodies in response to the parasite in the CNS.1,7
Natural transfer of antibodies from serum to CSF has been detected in various studies.8–11 Horses injected IM with a noninfectious protein develop specific antibodies in CSF after injection.10 Foals born to S neurona–seropositive mares have antibodies in the CSF after colostrum ingestion.8 In addition, anti–S neurona antibodies are detected in the CSF of horses vaccinated against S neurona that are apparently healthy and have negative test results in serum and CSF prior to vaccination.9,11 In 1 study9 that used the IFAT, a positive correlation between serum and CSF S neurona antibody titers was detected, indicating that the higher the serum antibody concentration the higher the likelihood of antibodies in CSF. In that study, serum and CSF titer-specific LRs were estimated and recommended for interpretation of test results.9 Titer-specific LRs allow interpretation of each test result individually and eliminate the need to establish a cutoff value for a positive test result, hence broadening the possibilities for test result interpretation.9,12,13
Findings of a correlation between serum and CSF antibodies in healthy horses raised the question of the value of CSF testing because naturally exposed horses might have anti–S neurona antibodies in the CSF in the absence of parasites in the CNS. Hence, CSF testing might not substantially increase diagnostic certainty, and misclassification of seropositive horses with neurologic signs that do not have EPM might occur. In addition, the possibility of use of LRs for specific test results when using quantitative tests such as the IFAT broadens the possibilities for interpretation of serologic test results alone, potentially decreasing the need for a CSF test. For instance, a horse with a high serum antibody titer is more likely to have antibodies in CSF, compared with a horse with a low serum antibody titer, even if it does not have EPM.9 Therefore, it might be advantageous for a clinician to make treatment decisions on the basis of clinical findings and serum test results alone instead of incurring the additional costs and risks associated with a spinal tap for CSF collection.
The objective of the study reported here was to assess the usefulness of CSF testing by use of the IFAT in the diagnosis of EPM caused by S neurona.
Materials and Methods
Data—Serum and CSF test results from a prior study evaluating the IFAT for anti–S neurona antibodies in naturally exposed, experimentally infected, or vaccinated horses were used in this study.9 Briefly, the IFAT was performed as described14,15 by use of S neurona merozoites (UCD-1 isolate) as the test antigen. Parasites were maintained in vitro, harvested by scraping, filtered to minimize cell debris, added to 12-well slides in 10-μL aliquots/well, and air dried.14Subsequently, antigen slides were fixed in 10% formalin for 10 minutes, double washed in PBSS, air dried, and stored at −70 °C for future use.14 Affinity-purified antibodies directed against horse-specific IgG and labeled with fluorescein were diluted 1:1,000 in PBSS and added to each well in 10-μL aliquots.14 Test results were reported as reciprocal IFAT titers from < 10 to 20,480 for serum and < 5 to 160 for CSF. The naturally exposed group comprised 122 horses necropsied at the California Animal Health and Food Safety Laboratory and categorized as S neurona infected or not on the basis of results of histologic and immunohistochemical examination of multiple sections of brain and spinal cord for detection of parasites or lesions compatible with protozoal infection.4,9 Serum, CSF samples, or both were collected from these horses at the time of necropsy. Experimentally infected horses were from 2 S neurona experimental infection trials.9,16,17 The first trial included 24 horses assigned to 5 treatment groups and 1 control group.9,17 Horses were subjected to stress of transport and subsequently inoculated orally with various doses (102 to 106) of S neurona sporocysts. Blood for serum samples was collected prior to sporocyst inoculation and then weekly. The CSF samples were collected prior to inoculation and at the end of the trial at approximately 30 days after infection. All 24 horses had serum and CSF IFAT titers < 10 and < 5, respectively, at the beginning of the trial.9,17 The second trial included 10 horses assigned to 2 treatment groups (4 horses/group) and 1 control group (2 horses).9,16 Horses in the treatment groups received 7 consecutive doses of 5 × 105S neurona sporocysts. Serum and CSF samples were collected at approximately 14-day intervals for 98 days. Serum IFAT titers prior to inoculation varied from < 10 to 40. Both control horses had serum IFAT titers ≤ 10 at the start of the trial. All 10 horses had CSF titers < 5 at the beginning of the trial.9,16 Clinical signs in experimentally infected horses ranged from equivocal to moderate.9,16,17At the end of both trials, horses were euthanized.9,16,17 Complete gross examination of the CNS was performed, followed by histologic examination of multiple sections of brain and spinal cord and, when appropriate, immunohistochemical evaluation.9,16,17 The vaccination trial included 20 vaccinated and 6 control horses.9The vaccination protocol consisted of two 1-mL doses of vaccinea (killed S neurona merozoites) administered IM 3 weeks apart. Serum and CSF samples were collected before the first vaccination and at 14, 28, and 112 days after the second vaccination.9 None of the horses in the vaccination trial developed clinical signs compatible with EPM.
One hundred eighty-two horses with 428 serum and 355 CSF test results were included in the present study. Test results from experimentally infected and vaccinated horses at different points in time after infection or vaccination were treated independently and considered to be representative of an individual horse's response to natural infection or vaccination at an equivalent time after infection or vaccination. Serum and CSF test results were assigned to 1 of 2 groups. The EPM+ group included serum and CSF test results from naturally and experimentally infected horses (≥ 8 days after infection) with parasites or lesions consistent with EPM in the CNS at the time of necropsy. One hundred two serum results, 89 CSF results, and 73 pairs of serum-CSF test results were assigned to this group. The EPM– group included serum and CSF test results from the naturally exposed and experimentally infected horses without parasites or lesions consistent with EPM in the CNS at the time of necropsy, experimentally infected horses prior to infection (0 days after infection), control horses in both experimental infection trials that did not have lesions compatible with EPM in the CNS at the time of necropsy, test results for the control horses from the first half (≤ 41 days after infection) of the experimental infection trials that had lesions compatible with EPM in the CNS, and all test results for the horses in the vaccination trial. In this group, there were 326 serum results, 266 CSF results, and 102 pairs of serum-CSF test results.
Model—Logarithmic transformation (natural logarithm) of the serum and CSF titers was performed. Probability distributions were fitted to the serum data in the EPM+ and EPM– groups separately. In each group, 2 probability distributions were selected to model the data and account for the uncertainty in the choice of the distribution. Several combinations of distributions were evaluated on the basis of plots overlaying the histogram of the actual data with the curve for the fitted distribution. The distributions that were considered for evaluation in each group were selected from lists of fitted distributions generated on the basis of best-fit statistics (χ2, Anderson-Darling, and KolmogorovSmirinov tests) by use of a specialized software package.b The final distributions were selected on the basis of visual evaluation of the plots and best-fit statistics. A discrete distribution was used to model the sampling from each of the 2 chosen distributions in each group (EPM+ and EPM–) with equal probability. Each distribution contributed 50% of the serum and CSF titer values in each of the 2 EPM groups.
The correlation between serum and CSF titers was modeled by use of the envelope method.18 Scatter plots that used the log-transformed serum and CSF titers were generated for each group (EPM+ and EPM–). On the basis of these plots, 2 regression lines representing the maximal and the minimal CSF titers for a given serum titer were generated. The highest and lowest serum titers with their respective highest and lowest CSF titers were used to generate the slope and intercept of the maximal and minimal regression lines. These bounding regression lines were used as parameters in a uniform distribution that modeled CSF titers for a specific serum titer. Therefore, for any given serum titer, a CSF titer was randomly selected from a uniform distribution with minimal and maximal parameters equal to the solution of the minimal and maximal regression lines equations for that specific serum titer.
On the basis of the distributions fitted to the data and in the serum-CSF correlation model, 4,000 iterations each generating 4,000 pairs of serum-CSF titers (total, 160,000 pairs) were obtained for the EPM+ and EPM– groups separately. The proportions of each serum, CSF, and serum-CSF pairs of titers for each of these 4,000 iterations were calculated. Titerspecific LRs for each serum, CSF, and serum-CSF pairs of titers were estimated by dividing the proportions of serum, CSF, and serum-CSF titers in the EPM+ group by the same proportions in the EPM– group. The number of iterations was determined on the basis of model convergence. Convergence was considered to have occurred when the mean and SD of the titer frequencies changed by < 2.5% as new iterations were run.
The post-test probabilities of EPM for serum IFAT titers (≤ 20, 40, 80, 160, and ≥ 320), CSF titers (< 5, 5, 10, and ≥ 20), and all 20 possible serum-CSF pairs of titers were estimated. These post-test probabilities were calculated on the basis of the relationship between post-test odds, pre-test odds, and LRs (post-test odds = pre-test odds × LR) as follows12:
where for each × serum, CSF, or serum-CSF pair of titer, Ps(X) is the post-test probability of EPM, LRX is the LR for a specific titer value (or combination of titers) X, and Pt is the pre-test probability of EPM (the pre-test probability for a CSF test after a serum test is equal to the post-test probability for the serum test result). Post-test probability refers to the likelihood of having EPM on the basis of a serum, CSF, or combination serum-CSF test result. Pre-test probability is the likelihood of having EPM before a serum or CSF testing.
The expected post-test probabilities of EPM for any of the possible CSF titers given a serum titer were calculated as follows:
where EPs is the expected post-test probability of EPM for possible CSF titers (< 5, 5, 10, ≥ 20) given a specific serum titer (≤ 20, 40, 80, 160, or ≥ 320), Ps is the post-test probability of EPM for the CSF titergiven the serum titer y calculated as in Equation 1 by use of the LR for the specific serum-CSF titer pair, and Po is the probability of occurrence of the CSF titer given the serum titer y calculated as follows:
where Po(xi/y) EPM+ and P(xi/y)EPMare the probabilities of a CSF titer xi given the serum titer y in the EPM+ and EPM– group, respectively, and Pt is the pre-test probability of EPM (before any testing).
Post-test probabilities and expected post-test probabilities of EPM were estimated for pre-test probabilities of EPM equal to 5%, 10% to 90%, and 95%. The pre-test probabilities of EPM for a serum test followed by a CSF test were updated according to the serum test results; these were equal to the post-test probabilities of EPM for a serum test only. All simulations were performed by use of a latin-hypercube sampling scheme via specialized software.b
Comparisons—For a serum test followed by a CSF test versus a serum test only, the absolute differences and proportional differences between the post-test and expected posttest probabilities of EPM were estimated over all values of pre-test probabilities. For a serum test only versus a CSF test only, the absolute and proportional differences between the post-test and the pre-test probability of EPM for each serum and CSF titer value alone were estimated. The expected differences and proportional differences were estimated and calculated as the product of each difference by its probability of occurrence. The probability of occurrence was estimated as in equation 3 by use of the probability of a serum or CSF titer alone in the EPM+ and EPM– groups. The difference between the expected differences for use of a serum test alone and CSF test alone were estimated across the range of pre-test probabilities of EPM.
Results
Model—The probability distributions fitted to the logarithmic-transformed IFAT serum titers in the EPM+ group were the Weibull (2.62, 4.92, Shift 1.12) and the Loglogistic (−17.1, 22.53, 217.76). The probability distributions fitted to the serum titers in the EPM– group were the Triangular (2.3, 2.3, 10.32) and the Exponential (2.49, Shift 2.28). The relative frequency, difference, and relative difference for the actual and the median simulated serum and CSF IFAT titers for the EPM+ and EPM– groups were determined (Table 1). The χ2 values for comparisons between actual and simulated frequencies for serum and CSF titers in both groups were ≤ 0.9 (P > 0.05). The actual and median simulated correlation coefficients between serum and CSF IFAT titers (logarithmic transformed) in the EPM+ group were 0.61 and 0.64, respectively. The actual and median simulated correlation coefficients between serum and CSF titers in the EPM– group were 0.49 and 0.44, respectively. The median, 5th, and 95th percentile for the simulated LRs was determined (Table 2).
Difference and relative difference between median simulated and actual serum and CSF indirect fluorescent antibody titer frequencies for EPM+ and EPM− groups of horses.
| Titer | EPM+ | EPM− | ||||||
|---|---|---|---|---|---|---|---|---|
| Actual (%) | Simulated (%) | Difference (%) | Relative difference (%) | Actual (%) | Simulated (%) | Difference (%) | Relative difference (%) | |
| Serum | ||||||||
| ≤ 20 | 28.4 | 23.9 | −4.5 | −15.9 | 75.8 | 70.1 | −5.7 | −7.5 |
| 40 | 4.9 | 7.5 | 2.6 | 52.5 | 4.0 | 5.5 | 1.5 | 36.7 |
| 80 | 11.8 | 10.2 | −1.6 | −13.3 | 3.1 | 4.9 | 1.8 | 59.8 |
| 160 | 11.8 | 11.9 | 0.1 | 1.2 | 4.0 | 4.4 | 0.4 | 9.1 |
| ≥ 320 | 43.1 | 46.5 | 3.4 | 7.9 | 13.2 | 15.2 | 2.0 | 15.3 |
| CSF | ||||||||
| < 5 | 36.0 | 27.2 | −8.7 | −24.3 | 89.5 | 88.4 | −1.1 | −1.2 |
| 5 | 14.6 | 22.2 | 7.6 | 52.0 | 6.0 | 7.0 | 0.9 | 15.6 |
| 10 | 14.6 | 11.1 | −3.5 | −24.2 | 2.6 | 3.0 | 0.4 | 15.0 |
| 20 | 34.8 | 39.5 | 4.7 | 13.4 | 1.9 | 1.6 | −0.3 | −13.5 |
Median, 5th, and 95th percentiles for simulated LRs for IFAT titers for serum, CSF, and combination of serum-CSF.
| Serum titer | CSF titer | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| < 5 | 5 | 10 | ≤ 20 | Serum only | ||||||
| Median | 5th–95th | Median | 5th–95th | Median | 5th–95th | Median | 5th–95th | Median | 5th–95th | |
| < 20 | 0.2 | 0.2–0.2 | 4.3 | 3.3–5.6 | 4.3 | 2.9–6.6 | 7.5 | 4.0–20.5 | 0.3 | 0.3–0.4 |
| 40 | 0.8 | 0.7–1.0 | 1.7 | 1.3–2.4 | 1.7 | 1.1–2.9 | 11.2 | 6.4–27.3 | 1.4 | 1.2–1.6 |
| 80 | 1 | 0.8–1.2 | 2.7 | 2–3.8 | 2.7 | 1.7–4.7 | 17.6 | 10–41.3 | 2.1 | 1.8–2.4 |
| 160 | 1 | 0.8–1.2 | 3.7 | 2.8–5.2 | 3.8 | 2.4–6.6 | 24.3 | 14.3–52.7 | 2.7 | 2.4–3.2 |
| ≥ 320 | 0.2 | 0.2–0.3 | 3.3 | 2.8–4 | 4.4 | 3.5–5.8 | 31 | 24.1–41.4 | 3.1 | 2.9–3.3 |
| CSF only | 0.3 | 0.3–0.32 | 3.2 | 2.9–3.6 | 3.7 | 3.1–4.3 | 24.4 | 20–30.5 | – | – |
Comparisons—Median simulated post-test and expected post-test probabilities of EPM, differences, and proportional differences for each serum titer for selected pre-test probability values were determined (Table 3). Median post-test probabilities of EPM for each CSF titer and each combination of serum-CSF titers with their respective probability of occurrence were also determined. Overall, the post-test probability of EPM increased as the serum and CSF titer increased. For any specific serum titer, the post-test probability of EPM increased as the CSF titer increased. The probability of occurrence of a specific serum-CSF titer combination decreased as CSF titers increased. For any specific serum titer, the probability of occurrence of CSF titers ≥ 5 increased as the pre-test probability increased.
Median simulated post-test probabilities of EPM, expected post-test probabilities, differences, and proportional differences for serum, CSF, and combinations of serum-CSF IFAT titer for selected pre-test probability values. Median probabilities of occurrence of each combination serum-CSF titer are in parentheses.
| Pre-test probability (%) | Serum titer | CSF titer | Serum only (%) | Expected (%) | Difference (%) | Proportionaldifference (%) | |||
|---|---|---|---|---|---|---|---|---|---|
| < 5 | 5 | 10 | ≥ 20 | ||||||
| 5 | ≤ 20 | 0.4* (96)† | 7.1 (3) | 7.1 (1) | 11.7 (0.4) | 1.8 | 0.7 | −1.0 | −59.3 |
| 40 | 5.5 (67) | 11.2 (20) | 11.0 (9) | 44.3 (4) | 6.7 | 8.6 | 1.9 | 28.1 | |
| 80 | 10.1 (66) | 22.9 (20) | 23.0 (9) | 64.2 (5) | 9.9 | 16.3 | 6.4 | 64.8 | |
| 160 | 12.3 (66) | 34.3 (20) | 34.3 (9) | 78.4 (6) | 12.6 | 22.4 | 9.9 | 78.8 | |
| ≥ 320 | 3.7 (63) | 34.7 (19) | 41.5 (8) | 83.4 (9) | 13.9 | 19.9 | 6.0 | 43.6 | |
| CSF only | 1.6 | 14.4 | 16.1 | 56.3 | – | – | – | – | |
| 10 | ≤ 20 | 0.9 (94) | 14.0 (4) | 13.9 (2) | 21.9 (1) | 3.6 | 1.7 | −2.0 | −53.8 |
| 40 | 10.8 (66) | 21.0 (20) | 20.6 (9) | 62.7 (5) | 13.2 | 16.4 | 3.1 | 23.8 | |
| 80 | 19.1 (65) | 38.6 (20) | 38.7 (9) | 79.1 (7) | 18.9 | 28.2 | 9.5 | 50.3 | |
| 160 | 22.8 (63) | 52.5 (20) | 52.4 (8) | 88.5 (8) | 23.3 | 36.4 | 13.1 | 56.3 | |
| ≥ 320 | 7.4 (60) | 52.9 (19) | 59.9 (9) | 91.4 (12) | 25.4 | 30.5 | 5.2 | 20.3 | |
| CSF only | 3.3 | 26.2 | 28.8 | 73.1 | – | – | – | – | |
| ≤ 20 | 1.9 (91) | 26.8 (6) | 26.7 (2) | 38.7 (1) | 7.8 | 4.3 | −3.5 | −45.1 | |
| 40 | 21.5 (63) | 37.4 (21) | 36.9 (9) | 79.1 (7) | 25.5 | 30.3 | 4.8 | 18.7 | |
| 80 | 34.7 (61) | 58.6 (21) | 58.7 (9) | 89.5 (9) | 34.4 | 46.3 | 12.1 | 35.4 | |
| 160 | 39.9 (59) | 71.3 (21) | 71.3 (9) | 94.5 (11) | 40.5 | 55.3 | 14.7 | 36.4 | |
| ≥ 320 | 15.3 (54) | 71.6 (19) | 77.1 (9) | 96.0 (17) | 43.4 | 45.7 | 2.4 | 5.6 | |
| CSF only | 7.2 | 44.3 | 47.6 | 85.9 | – | – | – | – | |
| 30 | ≤ 20 | 3.2 (88) | 38.5 (7) | 38.4 (3) | 52.0 (2) | 12.7 | 7.8 | −5.0 | −38.9 |
| 40 | 31.9 (60) | 50.6 (21) | 50.1 (9) | 86.6 (10) | 36.9 | 42.7 | 5.7 | 15.2 | |
| 80 | 47.7 (58) | 70.8 (21) | 70.9 (9) | 93.6 (12) | 47.3 | 59.7 | 12.6 | 26.7 | |
| 160 | 53.2 (55) | 81.0 (22) | 81.0 (9) | 96.7 (14) | 53.9 | 68.1 | 14.2 | 26.3 | |
| ≥ 320 | 23.7 (48) | 81.2 (20) | 85.2 (9) | 97.6 (23) | 56.8 | 57.6 | 0.9 | 1.6 | |
| CSF only | 11.7 | 57.7 | 60.9 | 91.3 | – | – | – | – | |
| 50 | ≤ 20 | 7.2 (81) | 59.4 (11) | 59.3 (5) | 71.7 (2) | 25.4 | 17.3 | −8.1 | −32.0 |
| 40 | 52.3 (54) | 70.5 (22) | 70.0 (10) | 93.8 (14) | 57.7 | 63.7 | 5.9 | 10.1 | |
| 80 | 68.0 (51) | 85.0 (22) | 85.0 (10) | 97.2 (17) | 67.7 | 78.2 | 10.6 | 15.7 | |
| 160 | 72.6 (46) | 90.9 (23) | 90.8 (10) | 98.6 (21) | 73.2 | 84.1 | 10.9 | 14.9 | |
| ≥ 320 | 42.0 (36) | 91.0 (20) | 93.1 (10) | 99.0 (34) | 75.4 | 76.3 | 1.0 | 1.3 | |
| CSF only | 23.6 | 76.1 | 78.4 | 96.1 | – | – | – | – | |
| 70 | ≤ 20 | 15.3 (75) | 77.3 (15) | 77.2 (7) | 85.5 (3) | 44.3 | 31.2 | −13.1 | −29.6 |
| 40 | 71.9 (48) | 84.8 (23) | 84.5 (10) | 97.2 (18) | 76.1 | 80.7 | 4.5 | 5.9 | |
| 80 | 83.2 (44) | 93.0 (23) | 93.0 (10) | 98.8 (23) | 83.0 | 89.9 | 7.0 | 8.4 | |
| 160 | 86.1 (37) | 95.9 (24) | 95.9 (11) | 99.4 (28) | 86.4 | 93.2 | 6.8 | 7.8 | |
| ≥ 320 | 62.8 (24) | 95.9 (20) | 96.9 (11) | 99.6 (45) | 87.8 | 89.8 | 2.1 | 2.4 | |
| CSF only | 41.8 | 88.1 | 89.5 | 98.3 | – | – | – | – | |
| ≤ 95 | 20 | 59.6 (67) | 96.5 (20) | 96.5 (9) | 98.0 (4) | 86.6 | 72.0 | −14.7 | −16.9 |
| 40 | 95.4 (41) | 97.8 (24) | 97.8 (11) | 99.7 (24) | 96.3 | 97.3 | 0.9 | 1.0 | |
| 80 | 97.6 (35) | 99.1 (25) | 99.1 (11) | 99.8 (29) | 97.6 | 98.8 | 1.2 | 1.3 | |
| 160 | 98.1 (26) | 99.5 (26) | 99.5 (11) | 99.9 (36) | 98.1 | 99.3 | 1.2 | 1.2 | |
| ≥ 320 | 93.2 (8) | 99.5 (21) | 99.6 (12) | 99.9 (59) | 98.3 | 99.3 | 0.9 | 1.0 | |
| CSF only | 85.4 | 98.4 | 98.6 | 99.8 | – | – | – | – | |
Percentage values. – = Not applicable
For pre-test probabilities of EPM between 5% and 95%, differences between the expected post-test probabilities for use of a serum and CSF test and the posttest probabilities for use of a serum test only were ≤ 19% in 95% of the simulations. The minimal and maximal differences were −20% and 21%, respectively. For all pre-test probability values evaluated, there was an increase in the post-test probability of EPM for use of a serum followed by a CSF test, compared with the use of a serum test alone, when serum titers were from 40 to ≥ 320. The largest increases and proportional increases (> 10%) in post-test probability were for serum titers between 40 and 160 and pre-test probabilities between 5% and 60%. Use of a CSF test when serum titers were ≥ 320 increased the post-test probability of EPM by a mean of ≤ 10%. There was a decrease in the post-test probability of EPM for the use of a serum followed by a CSF test, compared with a serum test alone, when serum titers were ≤ 20. The largest decreases (> 10%) were for pre-test probability values between 70% and 95%. The median difference between post-test probabilities of EPM for use of a serum followed by a CSF test (expected post-test probability) and use of a serum test alone for each IFAT titer value across pre-test probability values was determined (Figure 1).
Median differences between post-test probabilities of EPM for use of a serum test followed by a CSF test and use of a serum test only for indirect fluorescent antibody test titers from ≤ 20 to ≥ 320 and pre-test probabilities of EPM from 5% to 95%
Citation: American Journal of Veterinary Research 67, 5; 10.2460/ajvr.67.5.869
Median differences between post-test probabilities of EPM for use of a serum test followed by a CSF test and use of a serum test only for indirect fluorescent antibody test titers from ≤ 20 to ≥ 320 and pre-test probabilities of EPM from 5% to 95%
Citation: American Journal of Veterinary Research 67, 5; 10.2460/ajvr.67.5.869
Median differences between post-test probabilities of EPM for use of a serum test followed by a CSF test and use of a serum test only for indirect fluorescent antibody test titers from ≤ 20 to ≥ 320 and pre-test probabilities of EPM from 5% to 95%
Citation: American Journal of Veterinary Research 67, 5; 10.2460/ajvr.67.5.869
For all pre-test probabilities of EPM evaluated, 95% of the simulated absolute differences between the preand post-test probabilities for a serum test only and a CSF test only were ≤ 29% and ≤ 68%, respectively. The absolute change in the expected differences and proportional differences across all values of pretest probability of EPM for a serum and a CSF test only was determined (Figure 2). The change in post-test probability in relation to the pre-test probability of EPM was greater for a CSF test only, compared with a serum test only, in 100% of the simulations.
Expected difference and proportional difference between preand post-test probability of EPM for a serum test result or a CSF test result used alone for various pre-test probability values.
Citation: American Journal of Veterinary Research 67, 5; 10.2460/ajvr.67.5.869
Expected difference and proportional difference between preand post-test probability of EPM for a serum test result or a CSF test result used alone for various pre-test probability values.
Citation: American Journal of Veterinary Research 67, 5; 10.2460/ajvr.67.5.869
Expected difference and proportional difference between preand post-test probability of EPM for a serum test result or a CSF test result used alone for various pre-test probability values.
Citation: American Journal of Veterinary Research 67, 5; 10.2460/ajvr.67.5.869
Discussion
In general, the probability of a horse having EPM increased as IFAT serum titers, CSF titers, or both increased because higher serum and CSF titers were more frequent among infected than noninfected horses. Therefore, LRs increased as titer values increased. For any pre-test probability value, the post-test probability of EPM increased as the titer value increased. Also, for any specific serum, CSF, or serum-CSF titer combination, as the pre-test probability of EPM increased, the post-test probability of EPM increased. These findings were consistent with the relationship between pre-test probability, LRs, and post-test probability as described.9,12,13 This relationship is equivalent to that among prevalence, test sensitivity, test specificity, and predictive values.9,12,13 The term pre-test probability refers to the probability of disease prior to testing, which, in the present study, represented the prevalence of EPM among horses with specific clinical signs. Likelihood ratios represent how frequent specific test results are in the infected group, compared with the noninfected group, and encompass the concepts of sensitivity and specificity. The farther from 1 the LR for a certain test result is, the larger the frequency of that test result in the infected group, compared with the noninfected group (or vice versa for LRs < 1), and hence, less misclassification. The post-test probabilities represent the updated probabilities of disease given results of single or multiple tests and a specific pre-test probability. Therefore, they are analogous to result-specific predictive values.
The pre-test probability values used in this study were chosen to represent the prevalence of EPM in populations of neurologic horses with various clinical signs. The prevalence of EPM among horses with neurologic signs has been estimated at approximately 25% in 2 regions of the United States.4,19 The prevalence of EPM among horses with neurologic signs might vary by geographic region and is likely lower at present because of the widespread occurrence of West Nile virus infection among horses throughout the United States. However, it might be considered a reasonable initial estimate of the pre-test probability of EPM in a horse with nonspecific neurologic signs (eg, ataxia) and clinical history (mature horse, no history of trauma, and not vaccinated for West Nile virus). As additional clinical information becomes available and other neurologic diseases are included or excluded from the differential diagnostic list, the pre-test probability value decreases or increases accordingly.
On the basis of the results of the present study, addition of CSF testing when serum IFAT titers were from 40 to ≥ 320 increased the probability of EPM by a mean of < 20% in 95% of the simulations. The largest expected increases occurred when serum titers were 80 or 160 for pre-test probabilities between 5% and 50%, indicating that the largest expected increases in certainty in an EPM diagnosis by addition of a CSF test occurs in horses with low to moderate (5% to 50%) pre-test probabilities of EPM that had serum titers of 80 or 160. The lowest expected increases in probability of EPM by use of a serum-CSF test combination were found for serum titers ≥ 320, indicating that the contribution of a CSF test is limited when serum titers are high. There was a decrease in the post-test probability of EPM when the CSF of horses with serum titers ≤ 20 were tested, indicating that when serum titers are low, a CSF test will most likely help rule out EPM, especially as pre-test probabilities increase.
In the present study, the differences and proportional differences between the post-test probability of EPM for use of a serum-CSF test combination and use of a serum test alone represented mean values. These differences and proportional differences were calculated on the basis of the mean (expected) probability of EPM in a horse with a specific serum titer that could yield any CSF titer with a certain likelihood. Differences larger than the expected are possible; however, the largest differences were evident when CSF titers were > 5 and less likely to occur. The following example illustrates the interpretation of the data. Consider a horse referred for clinical examination with nonspecific neurologic signs (eg, ataxia) and clinical history (mature horse, no history of trauma, and not vaccinated against West Nile virus). In such a situation, EPM is one of the potential diagnoses and the pretest probability for the disease is equal to the prevalence of EPM in the population of neurologically abnormal horses of the practice, clinic, or hospital. On the basis of prior estimates of the prevalence of EPM among horses with neurologic signs in the United States,4,19 an initial pre-test probability of 20% will be considered. If this horse is tested by use of serum IFAT and has a titer of 80, the probability that this horse has EPM increases from the initial 20% (pre-test value) to 34% (post-test value). Whether a CSF test should follow to assist ruling in or ruling out EPM will depend on the expected increase or decrease in post-test probability achieved by using the test, taking into consideration any additional information about differential diagnosis that becomes available prior to the CSF collection. Exclusion or inclusion of other causes of disease in the differential list increases or decreases, respectively, the pre-test probability of EPM and affects the decisions about additional testing. As the pre-test probability of EPM increases or decreases to more extreme values, the contribution of a CSF test to the certainty of diagnosis decreases. In this example, assume that no further information is available at the time a decision must be made in relation to a CSF collection. The expected increase in this horse's probability of EPM achieved by addition of a CSF test to the diagnostic workup is approximately 12% (from 34% to 46%). However, because each individual CSF titer has a likelihood of occurrence that depends on the serum titer and the proportion of infected and noninfected horses in the population (pre-test probability of EPM), the most likely CSF test result for this horse is < 5 (61% chance of occurrence). This indicates that although a mean increase in probability of 12% might be expected, there will most likely be no or minimal change in the probability of EPM. The post-test probability of EPM is essentially the same in a horse with a serum titer of 80 and a CSF titer < 5 as in a horse with a serum titer of 80 without a CSF test. Notice that CSF titers ≥ 5 could considerably increase the post-test probability of EPM in relation to a serum titer alone and, ultimately, might alter clinical decisions; however, these titers are less likely to occur (39% chance of occurrence).
In the present study, a CSF test only was a better predictor of EPM, compared with a serum test only, corroborating findings of prior test validation studies.4,9Therefore, use of a CSF test alone might be an alternative to the use of a serum test only or the combination of serum-CSF testing. The risks associated with CSF collection were reported to be infrequent and consequences to be minimal.20–22 However, there have been no formal estimates of risk, and complications might occur during or after CSF collection. Complications associated with lumbosacral collections include physical harm to the horse and handler during the collection procedure, fibrous adhesion between the interacuate ligament and dura mater, extradural CSF leakage, diffuse subarachnoid hemorrhage, and hemorrhage in the spinal cord.20–22 Complications associated with atlantooccipital collections include herniation of the temporal cortex of the cerebral hemispheres under the tentorium cerebelli, herniation of the cerebellum through the foramen magnum, lacerations of the brain stem, and complications associated with general anesthesia and recovery.20,21 Septic meningitis or local tissue infection at the puncture site can also occur for both collection methods.20,21 In addition, blood contamination of the CSF during sample collection, especially from the lumbosacral site, is a common occurrence and might invalidate CSF test results.20–23 Therefore, use of a CSF test alone might be optimal for horses in which differential diagnostic procedures for other neurologic diseases facilitate or require a spinal tap (ie, cervical radiographs under general anesthesia or myelography). Higher post-test probabilities were obtained for a CSF test, compared with a serum test, when pre-test probabilities were from 30% to 70%. In such instances, the differences between preand post-test probabilities were the largest when a CSF test was used, compared with a serum test.
In this study, there was not a predetermined threshold difference in post-test probabilities considered relevant for a decision about whether to test CSF. The results were intended to provide the basis for decision making with the understanding that clinicians with different experiences might consider the changes in probabilities more or less relevant and, ultimately, should decide whether the risks associated with a spinal tap and the extra costs invariably associated with additional testing outweigh the potential gains in diagnostic certainty. The current (2005) approximate cost of a lumbosacral and an atlanto-occipital CSF collection and testing varies from $160 to $180 (US) and from $230 to $400 (US), respectively (Veterinary Medical Teaching Hospital, University of California, Davis, and Veterinary Teaching Hospital, Colorado State University). These costs include collection, sedation, and fees for injectable anesthesia, plus the costs of a CSF profile (total protein, RBC count, WBC count, and cytology) and an IFAT for S neurona. Presently, an S neurona serum or CSF IFAT costs approximately $30. On the basis of the experience of 2 of the authors (WDW, JTD), the frequency of complications associated with a spinal tap is < 5%.
The western blot test is presently the test most frequently used for diagnosis of EPM.1 This test yields dichotomous results (positive or negative) that essentially indicate the presence or absence of antibodies at an unknown concentration. Extrapolation of the results presented here to the western blot is problematic. A positive western blot result in serum might correspond to IFAT titers from < 10 to ≥ 320, making a decision on CSF testing based on serum western blot test results difficult. The IFAT is more specific than the western blot test for diagnosis of EPM caused by S neurona.14 At a cutoff titer of 80, sensitivity and specificity of the IFAT in serum are 83% and 97%, respectively.9 In CSF, sensitivity and specificity of the IFAT are 100% and 99%, respectively, with a cutoff titer of 5.9 A major advantage of the IFAT in relation to the western blot test is the possibility of use of titer-specific LRs.9,14Likelihood ratios allow for interpretation of each value of a test result, thereby minimizing loss of information inherent to dichotomous or dichotomized test results.9,12,13 On the basis of dichotomous test results, a test-positive or a test-negative horse is given the same clinical consideration in its respective category (positive or negative), regardless of the magnitude of the actual antibody concentration. The same is true for quantitative tests dichotomized on the basis of a predefined cutoff value. The results of this study indicate the benefits of use of titer-specific LRs for quantitative tests and their value for clinical decision making in relation to the most appropriate testing scheme (single or sequential testing).
The titer-specific LRs estimated in this study were lower than previous estimates that used a subset of the same IFAT results.9 This was attributed to the fact that IFAT titers at different times after infection or vaccination were treated independently and included in the EPM+ and EPM– groups. These IFAT titers at different times after infection or vaccination were considered representative of individual horses at an equivalent time after natural infection or vaccination. Therefore, the serum and CSF titer distribution in the infected and noninfected groups had a wider and more overlapping range of titers. In the EPM– group, inclusion of test results from experimentally infected and vaccinated horses increased the frequency of specific titer values, leading to lower and more conservative LR estimates. The serum and CSF titer distributions generated in this study were considered conservative and equivalent to mean titer distributions of populations of infected and noninfected horses with neurologic disease and various degrees of S neurona exposure, vaccination, or both. In the prior study, the group of S neurona–infected horses comprised naturally infected horses only with higher titer values, whereas the noninfected group included a few exposed horses and no vaccinated horses.9 Therefore, there was less overlap between the titer distributions of the infected and noninfected groups and larger LR estimates were obtained. The lower LR estimates obtained in the present study decreased the post-test probability estimates; however, the differences between post-test probabilities should remain essentially constant.
In the present study, the objective of modeling the data was to generate sufficient numbers of serum titers, CSF titers, and pairs of serum-CSF IFAT titers to estimate the titer-specific (or combination of titers) LRs and the post-test probabilities of EPM. Some of the combinations of serum-CSF titers in the actual data had no or insufficient numbers, making the estimation of LRs impossible or the estimates unstable. By use of this approach, distributions for LRs and post-test probabilities were generated, and the uncertainty around these estimates was considered. The choice of probability distributions to model the data as well as the regression lines to model the correlation serum-CSF by the envelope method18 was somewhat subjective. In addition, use of a uniform distribution for selection of CSF titer for a given serum titer was conservative. However, the comparisons between the actual and the simulated data indicated that the model reproduced the serum, CSF, and correlation serum-CSF data well and produced a good fit.
In horses and other domestic species, detection of antibodies in CSF has been used or suggested for the diagnosis of diseases such as West Nile virus infection,24 Togaviridae infections in horses,25,26 canine distemper,27 and feline infectious peritonitis,28 among others.29 In humans, detection of antibodies in CSF has been used or suggested for the diagnosis of viral, bacterial, fungal, protozoal, and parasitic neurologic diseases.30–32 Results of the present study indicate that sequential testing of serum and CSF for detection of antibodies against infectious agents in horses and potentially in other species, including humans, should be done cautiously. Gains in diagnostic certainty by addition of a CSF test to a serum test might be limited or nonexistent because of the dependence (correlation) between test results. Use of a CSF test only might be an alternative when other differential diagnostic procedures require a spinal tap. Ruling out other causes of neurologic disease reduces the necessity of further testing.
ABBREVIATIONS
| EPM | Equine protozoal myeloencephalitis |
| IFAT | Indirect fluorescent antibody test |
| LR | Likelihood ratio |
| PBSS | PBS solution |
| EPM+ | EPM positive |
| EPM− | EPM negative |
References
- 1.↑
Furr M, MacKay R & Granstrom D, et al. Clinical diagnosis of equine protozoal myeloencephalitis (EPM). J Vet Intern Med 2002;16:618–621.
- 2.↑
Bentz BG, Granstrom DE, Stamper S. Seroprevalence of antibodies to Sarcocystis neurona in horses residing in a county of southeastern Pennsylvania. J Am Vet Med Assoc 1997;210:517–518.
- 3.
Blythe LL, Granstrom DE & Hansen DE, et al. Seroprevalence of antibodies to Sarcocystis neurona in horses residing in Oregon. J Am Vet Med Assoc 1997;210:525–527.
- 4.
Daft BM, Barr BC & Gardner IA, et al. Sensitivity and specificity of western blot testing of cerebrospinal fluid and serum for diagnosis of equine protozoal myeloencephalitis in horses with and without neurologic abnormalities. J Am Vet Med Assoc 2002;221:1007–1013.
- 5.↑
Saville WJ, Reed SM & Granstrom DE, et al. Seroprevalence of antibodies to Sarcocystis neurona in horses residing in Ohio. J Am Vet Med Assoc 1997;210:519–524.
- 6.↑
Tillotson K, McCue PM & Granstrom DE, et al. Seroprevalence of antibodies to Sarcocystis neurona in horses residing in northern Colorado. J Equine Vet Sci 1999;19:122–126.
- 7.
MacKay RJ, Granstrom DE & Saville WJ, et al. Equine protozoal myeloencephalitis. Vet Clin North Am Equine Pract 2000;16:405–425.
- 8.↑
Cook AG, Maxwell VB & Donaldson LL, et al. Detection of antibodies against Sarcocystis neurona in cerebrospinal fluid from clinically normal neonatal foals. J Am Vet Med Assoc 2002;220:208–211.
- 9.↑
Duarte PC, Daft BM & Conrad PA, et al. Evaluation and comparison of an indirect fluorescent antibody test for detection of antibodies to Sarcocystis neurona, using serum and cerebrospinal fluid of naturally and experimentally infected, and vaccinated horses. J Parasitol 2004;90:379–386.
- 10.↑
Furr M. Antigen-specific antibodies in cerebrospinal fluid after intramuscular injection of ovalbumin in horses. J Vet Intern Med 2002;16:588–592.
- 11.
Witonsky S, Morrow JK & Leger C, et al. Sarcocystis neurona-specific immunoglobulin G in the serum and cerebrospinal fluid of horses administered S neurona vaccine. J Vet Intern Med 2004;18:98–103.
- 12.↑
Dujardin B, Van den Ende J, & Van Gompel A, et al. Likelihood ratios: a real improvement for clinical decision making? Eur J Epidemiol 1994;10:29–36.
- 13.
Simel DL, Samsa GP, Matchar DB. Likelihood ratios for continuous test results—making the clinicians' job easier or harder? J Clin Epidemiol 1993;46:85–93.
- 14.↑
Duarte PC, Daft BM & Conrad PA, et al. Comparison of a serum indirect fluorescent antibody test with two Western blot tests for the diagnosis of equine protozoal myeloencephalitis. J Vet Diagn Invest 2003;15:8–13.
- 15.
Marsh AE, Barr BC & Madigan J, et al. Sequence analysis and polymerase chain reaction amplification of small subunit ribosomal DNA from Sarcocystis neurona. Am J Vet Res 1996;57:975–981.
- 16.
Cutler TJ, MacKay RJ & Ginn PE, et al. Immunoconversion against Sarcocystis neurona in normal and dexamethasone-treated horses challenged with S. neurona sporocysts. Vet Parasitol 2001;95:197–210.
- 17.
Sofaly CD, Reed SM & Gordon JC, et al. Experimental induction of equine protozoan myeloencephalitis (EPM) in the horse: effect of Sarcocystis neurona sporocyst inoculation dose on the development of clinical neurologic disease. J Parasitol 2002;88:1164–1170.
- 18.↑
Vose D. Modelling dependencies. In: Vose D, ed. Risk analysis: a quantitative guide. 2nd ed. New York: Chichester, 2000;418.
- 19.
Hamir AN, Moser G, Rupprecht CE. A five year (1985–1989) retrospective study of equine neurological diseases with special reference to rabies. J Comp Pathol 1992;106:411–421.
- 20.
Green E, Constantinescu G, Kroll R. Equine cerebrospinal fluid: physiologic principles and collection techniques. Compend Contin Educ Pract Vet 1992;14:229–237.
- 21.
Johnson P, Constantinescu G. Collection of cerebrospinal fluid in horses. Equine Vet Educ 2000;12:7–14.
- 22.
Mayhew IG. Collection of cerebrospinal fluid from the horse. Cornell Vet 1975;65:500–511.
- 23.
Miller MM, Sweeney CR & Russell GE, et al. Effects of blood contamination of cerebrospinal fluid on western blot analysis for detection of antibodies against Sarcocystis neurona and on albumin quotient and immunoglobulin G index in horses. J Am Vet Med Assoc 1999;215:67–71.
- 25.
Grabner A, Fischer A. Symptomatology and diagnosis of Borna encephalitis of horses. A case analysis of the last 13 years. Tierarztl Prax 1991;19:68–73.
- 26.
Reed S, Andrews FM & Matthews HK, et al. Disorders of the neurologic system. In: Reed S, Bayly W, Sellon D, eds. Equine internal medicine. 2nd ed. St Louis: WB Saunders Co, 2004;599–656.
- 27.↑
Leisewitz AL, Carter A & van Vuuren M, et al. Canine distemper infections, with special reference to South Africa, with a review of the literature. J S Afr Vet Assoc 2001;72:127–136.
- 28.↑
Foley JE, Lapointe JM & Koblik P, et al. Diagnostic features of clinical neurologic feline infectious peritonitis. J Vet Intern Med 1998;12:415–423.
- 29.↑
Dewey CW. Encephalopathies: disorders of the brain. In: Dewey CW, ed. A practical guide to canine and feline neurology. Ames, Iowa: Iowa State University Press, 2003;99–178.
- 30.
Gould LH, Fikrig E. West Nile virus: a growing concern? J Clin Invest 2004;113:1102–1107.
- 31.
Victor M, Ropper AH. Infections of the nervous system (bacterial, fungal, spirochetal, parasitic) and sarcoid. In: Victor M, Rooper AH, eds. Adams and Victor's principles of neurology. 7th ed. New York: McGraw-Hill Co Inc, 2001;734–782.
- 32.
Miller J, Jubelt B & Prockop LD, et al. Infections of the nervous system. In: Rowland LP, ed. Merritt's neurology. 10th ed. Philadelphia: Lippincott Williams & Wilkins, 2000;103–212.
