Journal of Analytical Toxicology, Vol. 29, July/August 2005

I TechnicalNote

Evaluation of the Post-Rotational Nystagmus Test (PRN) in Determining Alcohol Intoxication

Majda Zorec Karlov~ek* and lo~e Bala~ic Institute for Forensic Medicine, Medical Faculty, University of Ljubljana, Korytkova 2, 1000 Ljubljana, Slovenia

Abstract[

This study evaluated the accuracy of the post-rotational nystagmus test (PRN) on the basis of the results of 1006 PRN tests performed at the Institute for Forensic Medicine in Ljubljana between 1998 and 2002 during standardized medical examinations in cases of suspected drunk driving. The evaluation of PRN test results with blood alcohol concentration (BAC) as a reference was based on classification into the following categories and characteristics: true positives (TP), true negatives (TN), false positives (FP), false negatives (FN), sensitivity, specificity, positive predictive values (PPV), negative predictive values (NPV), and accuracy. An optimal cut-off value of 10 s for post-rotational nystagmus time was chosen with the help of a receiver operating characteristic (ROC) curve for the BAC limit of 0.5 g/kg. The results of the decision analyses were: TP = 584, FP = 43, FN = 229, TN = 150, sensitivity = 0.718, specificity = 0.777, PPV-- 0.931, NPV-- 0.396, and accuracy -- 0.730. The area under the ROC curve (AUC) was 0.813. Based on the AUC, the post-rotational nystagmus test is a good test for predicting alcohol intoxication over 0.5 g/kg. As a part of the physician's examination, it contributes significantly to the description of the clinical state.

PRN is induced by suddenly stopping the rapid rotation of the body; large slow movements of the eyeballs occur in the direc- tion opposite the direction of rotation.

Some people may have pathological nystagmus as a conse- quence of trauma, stroke, and diseases affecting the vestibular system, which controls balance, coordination, and orientation. Involuntary jerking of the eye with small amplitudes after stop- ping the rotation is not an important indicator of alcohol in- toxication (3).

In addition to alcohol, some central nervous system (CNS) de- pressants also have an influence on PRN time.

To perform this test, the person is turned with open eyes in a tight circle 5 times in 10 s around his axis (i.e., while sitting on a swivel chair), and then the time from when the chair stops to when the person fixes his gaze on an object (e.g., a pencil or finger) that the physician holds at a distance of approximately 25 cm from his eyes is measured.

The aim of the study was to evaluate the ability of the PRN test to correctly classify those with BAC values above and below 0.5 g/kg.

Introduction

With a legal blood alcohol concentration (BAC) limit of 0.5 g/kg (since 1965), Slovenia's road traffic legislation has in- creasingly oriented the work of forensic toxicologists towards quality assurance of blood and breath alcohol analyses and in- terpretation of the results. In comparison, the results of physi- cians' examinations have been neglected.

The introduction of a standardized test battery for identi- fying suspected offenders driving under the influence (DUI) at the roadside raised our interest in a new evaluation of the post- rotational nystagmus (PRN) test, or Taschen test (1,2), which is part of the physician's examination for detecting alcohol intox- ication.

" Author to whom correspondence should be addressed. E-mail: majda.zorec-karlovsek@mf.u ni-lj.si.

Materials and Methods

The retrospective study included the results of 1006 PRN tests performed by 13 medical doctors trained and experienced in medical examination for DUI cases at the Institute for Forensic Medicine in Ljubljana in the years 1998 through 2002. Cases involving a combination of alcohol and drugs were ex- cluded. A duplicate BAC determination was made with two dif- ferent headspace gas chromatographic procedures, and the mean was the value used for forensic purposes and also for our study.

We assumed that the mean BAC represented the real BAC at the time of performing the PRN test. The time interval be- tween the PRN test and blood sampling ranged from at least 5 min to a maximum of 15 rain.

The evaluation of PRN test results with BAC as a reference was based on statistical methods including correlation analyses as well as decision analyses. The following categories and char-

390 Reproduction (photocopying) of editorial content of this journal is prohibited without publisher's permission.

Journal of Analyt ical Toxicology, Vol. 29, July/August 2005

acteristics are determined and/or calculated: true positives (TP), true negatives (TN), false positives (FP), false negatives (FN), sensitivity, specificity, positive predictive values (PPV), negative predictive values (NPV), and accuracy for different cut-off values

of the PRN test. With the help of a receiver operating charac- teristic curve (ROC curve) for BAC limit values (0.5 g/kg as well as 0.8 and 1.0 g/kg), the optimal cut-off values for PRN time were selected (4).

Dis t r ibu t ion of PRN va lues

120 100

20 0

10 19 28 37

P R N (s)

[=mBAC < 0 .5g /kg . .BAC>0.5g /kg ]

Figure 1. PRN time distribution of subjects tested (N = 1006).

50

45

40

35

3o ~ 25 ~' 20

15

10

5

0 , 0.00 0.50 1.00 1.50 2.00 2.50

BAC (g/kg)

Figure 2. Correlation between PRN time and BAC.

3.00 3.50

Table I. Decision Data at 0.5 g/kg BAC Limit Value and Varying Cut-Off Values for PRN Time (N-- 1006)*

PRN Cut-off

(s) TP TN FP FN Sensitivity Specificity PPV NPV

Results and Discussion

6 748 88 105 65 0.920 0.456 0.878 0.575 7 719 107 86 94 0.884 0.554 0.893 0.532 8 675 121 72 138 0.830 0,627 0.904 0.467 9 656 131 62 157 0.807 0.679 0.914 0.455

10 584 150 43 229 0,718 0.777 0.931 0.396 11 570 153 40 243 0.701 0.793 0.934 0.386 12 508 164 29 305 0.625 0.850 0.946 0.350 13 473 168 25 340 0.582 0.871 0,950 0.331 14 433 170 23 380 0.533 0.881 0,950 0.309 15 332 181 12 481 0.408 0.939 0.965 0.273 16 293 182 11 520 0.360 0.943 0.964 0.260

* Sensitivity = TP/(TP + FN); Specificity = TN/(FP+TN); Positive predictive values (PPV) = TP/(TP + FP); Negative predictive values (NPV) = TN/(FN + TN); and Accuracy = (TP + TN)/alI.

Males represented 92.25% of all subjects tested. The average age in the group was 35.5 years (range 14 to 83 years). BAC values were between 0.00 and 3.16 g/kg, mean 1.12 g/kg, median 1.14 g/kg. The distribution of BAC values was not Gaussian. Of those tested, 10.93% of persons were sober, that is, with BAC values lower than 0.1 g/kg. BAC values equal to or lower than the administrative limit of 0.5 g/kg were found in 183 subjects. The prevalence of intoxicated persons (BAC values above 0.5 g/kg) was 0.808.

Measured values of post-rotational nystagmus time ranged from i to 45 s, mean value 14 s, median 13 s. The distribution of PRN values is presented in Figure 1. The values of PRN time and BAC show a positive correlation (Figure 2) that is statisti- cally significant (r -- 0.54, p < 0.01). For the regression line, the following formula was obtained: PRN = 6.0 x BAC + 7.25.

The distributions of PRN groups with BAC < 0.5 g/kg and BAC > 0.5 g/kg overlap (Figure 1). Low PRN times could be seen in the group with BAC below 0.5 g/kg and also in alcohol-in- toxicated persons; high PRN times could also be found among sober persons. Looking for the optimal cut-off value of the PRN test, different cut-off values from 6 to 16 s were selected and, with the help of decision matrices, the values for sensitivity, specificity, and accuracy were calculated (Table I).

The ROC curve is a plot of the true positive rate (sensitivity) against the FP rate (1 - specificity) for the different possible threshold values of the diagnostic test (Figure 3).

Raising the threshold values for PRN time (from 6 to 16 s) shows an increase in sensitivity accompanied by a decrease in

specificity. The competition between sensitivity and specificity leads to the estimation of the op- timal cut-off value of the PRN test, which ac- cording to the shape of the ROC curve is 10 s.

According to previous studies summarized by Heifer and colleagues (3,5), PRN times of

Accuracy healthy, sober individuals are as long as 10 s. This is in agreement with the selected PRN cut-

0.831 off value in our study for the BAC limit value of 0.821 0.5 g/kg.

0.791 The area under the ROC curve (AUC) is 0.813. 0,782 A rough guide for classifying the accuracy of a 0.730 0.719 diagnostic test is the traditional academic point 0.668 system: 0.90-1.0 = excellent (A), 0.80-0.90 = 0.637 good (B), 0.70-0.80 = fair (C), 0.60-0.70 = poor 0.599 (D), and 0.50-0.60 = fail (F). In line with this 0.510 classification, the PRN test can be classified as a 0.472 "good" one.

The introduction of higher BAC limits and higher BAC cut-off levels (0.8 g/kg and 1.0 g/kg) reduces the prevalence of intoxicated persons

391

R O C curve for P R N

0 0.2 0.4 0.6 0.8 1 1 - Specif icity

�9 B A C > 0 . S g / k g [

Figure 3. ROC curve for PRN cut-offs from 6 to 16 s and BAC limit of 0.5 g/kg,

Journal of Analyt ical Toxicology, Vol. 29, July/August 2005

in the group observed, followed by an increase of TN and FP, raising the sensitivity and lowering the specificity of the test. The consequences can be seen in the changes of the shape of the ROC curve and also in the changes of the AUC.

For a BAC limit value of 0.8 g/kg, the AUC is 0.809, and the optimal PRN time 11 s, and for a BAC limit value of 1.0 g/kg, the AUC is 0.785, and the optimal PRN time 12 s.

Table II contains some results of the study of age on the sta- tistical values of the PRN test. It was expected that the 30-34, 35-39, and 40--44 year age groups would be connected with the highest average BAC and the highest prevalence of intoxicated participants. The 40-44 year age group had the highest av- erage BAC at 1.30 g/kg, followed by the 35-39 and 30-34 year groups with a mean BAC value of 1.27 g/kg. The same groups also had the highest prevalence of BAC values over 0.5 g/kg, which varied from 0.842 to 0.857.

The age group "younger than 20 years" (n = 34) had an av- erage BAC of 0.61 g/kg, and for statistical purposes was con- nected to the group with 176 participants with an age range from 20 to 24 years and an average BAC of 0.94 g/kg. This yielded a group with 210 participants (age < 24 years) and a BAC of 0.72 g/kg. Because of the small number of older participants,

Table II. Decision Data at 0.5 g/kg BAC Limit Value, PRN Test Cut-Off at 10 s, and Varying Age of Participants

Years r* Mean BAC n TP TN FP FN Sensitivity Specificity PPV NPV Accuracy Prevalence

< 24 0.505 0,72 210 101 45 11 53 0.656 0.804 0.902 0 .459 0,695 0,733 24-29 0,465 1,04 188 102 21 9 56 0.646 0.700 0.919 0 .273 0.654 0,840 30-34 0.534 1,27 122 78 15 2 27 0.743 0.882 0.975 0 .357 0,762 0.861 35-39 0,583 1,27 133 89 18 3 23 0.795 0.857 0.967 0 .439 0,805 0.842 40-44 0.537 1.30 105 68 14 1 22 0.756 0.933 0.986 0 .389 0.781 0.857 45-49 0,584 1.20 119 73 17 5 24 0,753 0.773 0.936 0 .415 0.756 0.815 >50 0.521 1.13 129 73 20 12 24 0.753 0.625 0.859 0 .455 0.721 0.752

* r = correlation coefficient; n = number of participants in the age group; Sensitivity = TP/(TP + FN); Specificity = TN/(FP + TN); Negative predictive values (NPV) = TN/(FN + TN); and Accuracy = (TP + TN)/alI.

Positive predictive values (PPV) = TP/(TP + FP);

Table III. Decision Data at 0.5 g/kg BAC Limit Value, PRN Test Cut-Off at 10 s, and Varying Observers

No.* n Mean BAC Meanage r TP TN FP FN Sensitivity Specificity PPV NPV Accuracy Prevalence

1 52 1.08 36.85 0.820 20 10 1 21 0.488 0,909 1.000 0 .323 0.577 0.789 2 55 1.11 36.92 0.712 18 13 1 23 0.439 0.929 0.849 0.361 0.564 0.746 3 123 1,15 35.06 0.711 77 18 3 25 0.755 0.857 0.963 0 .419 0.772 0.829 4 167 1.01 33,97 0.705 82 39 6 40 0,672 0.867 0.932 0 .494 0.724 0.731 5 92 1,18 33.98 0,658 49 11 2 30 0.620 0.846 0.961 0 .268 0.652 0.859 6 122 1.09 36.64 0.616 78 17 8 19 0.804 0.680 0 .907 0 .472 0.779 0,795 7 17 1.20 33.00 0.592 9 3 0 5 0.643 1,000 1,000 0 .375 0.706 0,824 8 91 1.10 35.90 0.559 55 17 3 16 0.775 0,850 0,948 0,510 0.791 0.780 9 82 1.33 36.24 0.534 47 5 0 30 0.610 1,000 0.947 0 .143 0.634 0.939

10 22 1.15 31.86 0.534 8 4 0 10 0.444 1.000 1.000 0 ,286 0.545 0.818 11 24 0.87 34.88 0.530 14 4 5 1 0.933 0.444 0.737 0.800 0.750 0.625 12 63 1,03 37.55 0.509 45 5 8 5 0,900 0.385 1.000 0 .500 0.794 0.794 13 96 1.19 36.02 0.501 82 4 6 4 0.954 0.400 0.932 0 .500 0.895 0.896

* No. = the code of the observer (medical doctor); n = number of participants in the group; r= correlation coefficient; Sensitivity = TP/(TP + FN); Specificity = TN/(FP + TN); Positive predictive values (PPV) = TP/(TP + FP); Negative predictive values (NPV) = TN/(FN + TN); and Accuracy = (TP + TN)/alI.

392

Journal of Analytical Toxicology, Vol. 29, July/August 2005

only one group was formed (> 50 years) with 129 participants and an average BAC of 1.13 g/kg. The group of participants with an age > 50 years had a higher average BAC and a higher prevalence of BAC values over 0.5 g/kg than the youngest group (under 24 years), but lower values than the middle-aged groups.

Examination of the correlation coefficients between the PRN and the BAC values indicates that the PRN values correlate weakly to moderately with BAC in most subgroups. The highest correlation coefficient (r = 0.584) was achieved in the 45-49 age group, and the lowest (r = 0.465) in the 24-29 year group. BAC values over 0.5 g/kg were predicted with the help of the PRN test (cut-off = 10 s) with the highest accuracy (0.805) and the highest sensitivity (0.795) in the 35-39 year group. The highest specificity (0.933) and positive predictive power (0.986) of the test was found in the 40-44 year group.

In addition to age, the drinking habits, tolerance to alcohol, and state of health of the participants played an important role in the facts observed.

In spite of a prescribed procedure, the results of the PRN test depend on the observer. The factors influencing the result are the real rotation frequency, the response time of the ob- server, the observer's subjective decision about the amplitude, and the end of involuntary jerking of the eyeballs.

The number of participants in the group examined by one medical doctor, the average age, the average BAC, the preva- lence of participants with a BAC over 0.5 g/kg, correlation co- efficients, and decision data are summarized in Table III. The number of participants in the groups ranged from 17 to 167. The average BAC in the groups varied from 0.87 to 1.20 g/kg, and the average age varied from 31.86 to 37.55 years.

The results of work by 13 different physicians, identified with numbers from I to 13, showed a moderate-to-strong correlation between PRN and the BAC. The correlation coefficient was from 0.501 to 0.820. The PRN test was performed by different med- ical doctors with a sensitivity from 0.439 to 0.954 and a speci- ficity from 0.385 to 1.000. The diagnostic accuracy of the PRN test in predicting BAC values over 0.5 g/kg varied from 0.545 to 0.895. The BAC and PRN data collected by medical doctors No. 1 and No. 2 had a high correlation coefficient. On the other hand, the chosen PRN cut-off value of 10 s was too high for their performance of the test, which resulted in a high percentage of FN, low sensitivity, and low accuracy of the test. The observed inter-observer variability highlighted the need for the automa-

tization of the test performance entirely or, in some cases, the determination of observer-dependent PRN cut-off values

The PRN test is not used as a screening test at the roadside. The important role that it has as a part of the clinical examina- tion followed by blood alcohol determination is in providing valuable information about individual effects of alcohol. Driving under the influence of alcohol in Slovenia is a misdemeanor, and the penalty is based on the BAC alone. However, the de- scription of the state "under the influence of alcohol" in legal cases connected with traffic accidents and other criminal of- fences is an integral assessment of the factual system com- bining information acquired from statements made by the po- lice, results of a medical examination, results of toxicological examinations, and information obtained on the person under investigation.

Conclusions

The values of PRN time and BAC show a positive correlation that is statistically significant (r = 0.54, p < 0.01). The PRN test with a cut-off value of 10 s performed in line with the stan- dardized procedure has a sensitivity of 0.718, specificity of 0.777, and accuracy of 0.730. The PRN test is considered to be "good" at separating subjects with a BAC level below and above 0.5 g/kg. The area under the ROC curve is 0.813.

References

1. R. Manz. Der Einflug geringer Alkoholgaben auf die Teilfunk- tionen von Auge und Ohr. Dtsch. Zf.gerichtl. Med. 32" 301 (1939).

2. B. Taschen. Nystagmuspr~fungen zur Feststellung der Alkohol- beeinflussung. Dtsch. Z.f.gerichtl. Med. 44:436 (1955).

3. U. Heifer and D.Wichmann. Der grobschl~gige Drehnachnys- tagmus als Zeichen der Alkoholwirkung. Blutalkohol 1:257-271 (1962).

4. H.M. Zweig and G. Campbell. Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. Clin. Chem. 39:561-577 (1993).

5. U. Heifer, K. Selliger, and M. Kutzner. Experimentelle und statis- tische Untersuchungen 0ber den alkoholbedingten postrota- torischen Fixationsnystagmus. Blutalkohol 3:537-557 (1966).

393