The following expert report from Willett v. Teva explains the link between use of Reglan (or Metoclopramide) and movement disorders such as Tardive Dyskinesia.
Dear Mr. Huston:
The purpose of this letter is to meet the disclosure requirements for my testimony as an expert witness in the above civil case. Specifically, I will express my views, based on epidemiologic evidence, about the possible effect of metoclopramide use, especially for more than 12 weeks, on the incidence of tardive dyskinesia in adults (i.e., whether metoclopramide is a risk factor for tardive dyskinesia) and about the magnitude of tardive-dyskinesia risk among long-term users of metoclopramide (i.e., the probability of developing tardive dyskinesia after so many years of use).
I have three professional degrees in architecture (B.Arch., MIT, 1969), city planning (M.R.P., UNC-Chapel Hill, 1974), and epidemiology (Ph.D., UNC-Chapel Hill, 1978). For the past 30 years, I have been practicing and teaching epidemiology with faculty appointments at the Yale University School of Medicine (1978-85), the UCLA School of Public Health (1985-2003), and the University of Michigan (UM) School of Public Health (11/03 to present). Currently, I am the Chair of the Department of Epidemiology at UM, Professor of Epidemiology and Environmental Health Sciences, and Director of the Graduate Summer Session in Epidemiology at UM. I have extensive research experience in a wide range of public-health areas, including neuropsychiatric disorders, musculoskeletal conditions, cancers, cardiovascular disease, nonintentional injuries, research methods, occupational and environmental health, psychosocial aspects of disease, and access to and quality of health care. I have conducted several studies of tardive dyskinesia, including investigations of drug effects, and I have co-authored numerous papers from this research that were published in peer-reviewed journals (refer to my CV, which lists all my publications). In addition, I have received four teaching awards in the schools of public health at Yale (1985) and UCLA (1989; 1997; and 1999) and the Tyroler Distinguished Alumni Award from the University of North Carolina at Chapel Hill, Department of Epidemiology (2003).
My fee for all background work is $450 per hour, and my fee for deposition and trial testimony is $550 per hour. All of my testimony in the past five years is listed in Appendix A.
Tardive Dyskinesia (TD) and Metoclopramide (MET)
TD is an involuntary movement disorder of the orofacial region, trunk, and extremities, which is believed to be caused in part by the long-term use (> 1-3 months) of antipsychotic medications. There have been numerous epidemiologic studies of TD among patients maintained on antipsychotics, and we know that factors other than antipsychotic drug use appear to increase the risk of TD, including older age.
MET is a dopamine receptor antagonist that has been used as a prokinetic and antiemetic agent to treat several disorders, including diabetic gastroparesis, esophageal reflux, and chemotherapy-induced nausea. MET has antidopaminergic properties similar to conventional antipsychotic drugs that are known to cause TD, and it has been shown to induce dyskinetic-like movements (oral stereotypies) in pigs (Jeste & Wyatt, 1982). Consequently, the effect of MET on TD in humans is biologically plausible, especially when used for long periods. Indeed, the possible effect of MET has been recognized by researchers and clinicians for more than 25 years (Grimes et al., 1982; Jeste & Wyatt, 1982; Sewell & Jeste, 1992). Despite the recommendation that MET be used for no more than 12 weeks, it is frequently prescribed off-label for much longer periods (Pasricha et al., 2006). Unfortunately, there seems to be little empirical research documenting the amount of off-label use in the U.S. In the only relevant study I could find, Stewart et al. (1992) observed 4,515 elderly participants in an ambulatory screening program between 1978 and 1979. Among the 34 subjects who were using MET, 11(32%) reported use for more than one year. Following the marketing in 1993 of cisapride, a prokinetic drug that does not appear to produce extrapyramidal symptoms, the utilization of metoclopramide decreased appreciably in the U.S. When cisapride was subsequently linked with serious cardiac arrhythmias, it was withdrawn from the market in 2000, and the utilization of MET increased to its pre-cisapride level (Shaffer et al., 2004).
Measures of Disease Frequency
Four measures for quantifying the frequency of tardive dyskinesia (TD) in a population are used in my testimony and reports: incidence rate; risk; point prevalence; and period prevalence. The first two measures represent different methods for quantifying the occurrence of new (incident) TD cases among persons at risk of the disease; the latter two measures represent different methods for quantifying the number of existing (prevalent) TD cases in a population.
The incidence rate is an instantaneous measure that can change from moment to moment (like the velocity of a moving vehicle). The average incidence rate for a given follow-up period is computed as the number of new cases occurring during the period, divided by the amount of person-time experienced by the population at risk during that period (where, e.g., 10 persons at risk followed for two years accumulate 10×2 = 20 person-years). Thus, incidence rates are not dimensionless quantities; they have units of 1/time. For example, if 12 new cases occur in 1,000 person-years of follow-up, the incidence rate is 0.012/year.
Risk (or cumulative incidence) is a cumulative measure that reflects the probability of individuals developing the disease during a given follow-up period. The average risk for a given period in a population may be computed as the number of new cases occurring during the period, divided by the number of persons at risk at the start of the follow-up period (provided that all noncases are followed for the entire period). Thus, risks do not have units of measurement (as do rates), but they do have a specific time referent, e.g., the 1-year risk or the 5-year risk. For example, if 100 persons at risk are followed (without loss) for two years and 15 new cases of TD occur during that period, the 2-year risk is 0.15 or 15%.
Since both the rate and risk reflect the incidence of disease in a population, the two measures are related mathematically. Specifically, if the incidence rate (IR) remains constant during follow-up, the A risk (for a follow-up period of duration delta) is equal to 1-exp[-[-RxA], where “exp” is the natural antilog of the quantity in brackets, and IR and A are expressed in complementary units (e.g., per year and years). For example, if the constant incidence rate of TD is 0.05/year, the 5-year risk is 1-exp.[-0.05×5] = 22.1%.
Point prevalence (or simply prevalence) is a cross-sectional measure that reflects the probability of someone in a population being a case at a given (point in) time. (Note that prevalence is not a rate.) The prevalence of disease in a population is computed as the number of existing (prevalent) cases at a given time, divided by the total number of persons in that population at that time. The prevalence of disease in a population depends on the incidence rate of that disease as well as the mean duration of the disease among cases (i.e., the course of the disease) and selective migration of cases and noncases into and out of the population. Thus, a low prevalence may reflect a low incidence rate of the disease, a short duration of the disease (due to quick recovery or death), and/or selective outmigration of cases from the population.
Period prevalence is another cross-sectional measure that reflects the probability of someone in a population being or having been a case at any time during a given period. The A period prevalence (for a previous period of duration delta) in a population at the end of that period is computed as the number of persons who have or had the disease any time during that previous period, divided by the total number of persons in the population. To the extent that persons with the disease can recover, period prevalence will be greater than point prevalence in a particular population; the longer the period (A), the greater the difference between the expected point and period prevalence.
Measures of Statistical Association
To estimate the effect of an exposure such as drug use on the occurrence of a disease, epidemiologists calculate measures of the statistical association between exposure status and disease status in the population. This is typically done by estimating the ratio of disease frequency, comparing exposed persons with unexposed persons, where disease frequency is measured as the incidence rate, risk, or prevalence. Thus, three measures of association are called the incidence rate ratio (or simply rate ratio), the risk ratio (or relative risk), and the prevalence ratio.
In certain types of epidemiologic studies – namely, case-control and cross-sectional studies (like those used in my testimony) – the measure of association typically used to estimate effect is the odds ratio. In a cross-sectional study of disease prevalence, for example, the prevalence odd ratio is the odds of being a case in the exposed group divided by the odds of being a case in the unexposed group, where the odds of being a case in each group is disease prevalence (P) divided by (1-P). In the absence of epidemiologic bias, which is discussed later in this report, the prevalence odds ratio may be interpreted as the incidence rate ratio.
Estimating the Effect of MET on TD
The first part of my testimony deals with the effect of MET use for more than 12 weeks on the occurrence of TD. To the best of my knowledge, there are three epidemiologic studies in which the investigators estimated the effect of MET on TD in humans by comparing users and nonusers of MET:
Ganzini L, Casey DE, Hoffman WF, McCall AL. The prevalence of metoclopramide-induced tardive dyskinesia and acute extrapyramidal movement disorders. Archives of Internal Medicine 1993; 153:1469-75.
Sewell DD, Kodsi AB, Caligiuri MP, Jeste JV. Metoclopramide and tardive dyskinesia. Biological Psychiatry 1994; 36:630-32.
Matson JL, Mayville EA, Bielecki J, Smalls Y, Eckholdt CS. Tardive dyskinesia associated with metoclopramide in persons with developmental disabilities. Research in Developmental Disabilities 2002; 23:224-233.
There are also several “case-series” investigations (e.g., see Sewell & Jeste, 1992), but the results of these studies cannot be used to estimate the effect of MET on TD because the investigators do not have information on the population from which the TD cases were identified. Thus, they cannot be used to measure the statistical association between MET use and TD status.
Before reviewing each of these studies, I would like to point out several similarities in the methods and study populations of the first two studies by Ganzini et al. (1993) and Sewell et al. (1994). First, both studies are cross-sectional designs in which the estimated prevalence (presence of existing cases) of TD was compared for persons treated with MET and persons not treated with MET. Second, the subjects in both studies were veterans – mostly white males with mean ages of 60-64 years – who were being treated at VA medical centers (but in different states). Third, the investigators of these two studies used the same exam-based scale (the Abnormal Involuntary Movement Scale, AIMS) to detect dyskinetic (TD) movements and similar symptom-based criteria to identify TD cases (Schooler-Kane criteria [Schooler & Kane, 1982]; and both sets of investigators “blinded” the TD raters (examiners) to MET treatment status and other diagnoses. In addition, patients with any previous use of antipsychotics or with other neurologic diseases (aside from TD) known to cause movement disorders were excluded from both studies.
To measure the association between MET and TD, I used published results from Ganzini Page 5 et al. (1993) and Sewell et al. (1994) to calculate the prevalence odds ratio (defined above). As already noted, in the absence of epidemiologic bias, the prevalence odds ratio (hereafter called the “odds ratio”) may be interpreted as the incidence rate ratio – i.e., the incidence rate of TD in MET users divided by the incidence rate in nonusers (Rothman & Greenland, 1998:64). This incidence rate ratio (or simply “rate ratio”) is an appropriate parameter for measuring the net effect of MET use on the occurrence of TD in the population. Furthermore, except when the prevalence is very low in both exposure groups, which is not true in either study, the odds ratio is generally a better estimate of the incidence rate ratio than is the prevalence ratio (i.e., the prevalence of TD in MET users divided by the prevalence in nonusers). Nevertheless, since Ganzini et al. based their analysis on prevalence ratios, I have conducted the combined analysis using both measures. In general, the prevalence ratio is closer to the null value (1, indicating no association) than is the odds ratio.
Ganzini et al. (1993): Fifty-one (51) nonusers of MET, who were general medical and endocrinology patients, were individually matched to 51 MET users on age, gender, and diabetes status. Matching on diabetes status was done because these investigators had previously found an association between diabetes and TD among users of both conventional antipsychotics and MET (Ganzini et al., 1991; 1992). All users had been receiving MET for at least three months (mean dose = 31 mg/day), and the average duration of previous use was 2.6 years. The estimated odds ratio (OR) was 1.94 (95% confidence interval [CI] = 0.76, 4.97), and the two-sided p value for testing the null hypothesis of no association (OR = 1) was 0.083. The authors report that users and nonusers of MET were similar in the distributions of other known or possible risk factors for TD – specifically, the use of anticholinergic medications, the presence of an affective disorder, and the three matching variables. Thus, it appears that the estimated OR was not appreciably confounded by these other factors.
As noted above, Ganzini et al. used the prevalence ratio (PR) to measure the association between MET and TD. Using the information in Table 1 of their paper, I derived the results needed for a matched analysis. There were 15 matched pairs in which the MET user had TD, 9 matched pairs in which the nonuser had TD, and 6 matched pairs in which both subjects had TD. Thus, the estimated PR was 15/9 = 1.67 (95% CI = 0.93, 2.99). The p value was the same as above, 0.083.
Sewell et al. (1994): Fifty-one (51) users of MET and 34 nonusers were selected from the same clinics and inpatient units of a VA medical center. All users had been receiving MET for at least 30 days. Although the authors do not report the mean duration of use, I would expect it to be much greater than three months. The estimated OR was 2.84 (95% CI = 0.85, 9.53; p = 0.085). While formal matching was not used in the selection of subjects, the authors report that users and nonusers of MET were similar in the distributions of several known and possible risk factors for TD, including age, gender, race, and diabetes status. Thus, it appears that the estimated OR was not appreciably confounded by these other factors.
Using the prevalence ratio to measure the MET-TD association, we get PR = 2.33 (95% CI = 0.84, 6.49), and the p value is 0.085 (same as above).
Matson etal. (2002): Seventy-five (75) adults with mental retardation and a mean age of 40 years were examined for TD symptoms every three months for one year, using the Dyskinesia Identification System – Condensed User Scale (DISCUS). Mean DISCUS scores and TD prevalence were compared at each exam and across exams (period prevalence) among three groups of equal size that were matched on age, gender, and severity of mental retardation: 1) subjects treated with MET (mean dose = 33 mg/day); 2) subjects treated with conventional antipsychotics (mean chlorpromazine-equivalent dose =211 mg/day); and 3) subjects not treated with either type of drug (examined only at the end of follow-up). Users of both drugs had been treated with those medications for at least 4 years at baseline, and doses were stable during follow-up. (Note: the authors' claim that they were estimating “incidence rates” of TD is incorrect, because they did not analyze changes in TD status, i.e., from being a noncase to being a case). At the last exam, the mean DISCUS score was 2.5 in the antipsychotic group, 2.3 in the MET group, and 1.0 in the untreated group (p value for testing the null hypothesis that the mean was the same in all groups = 0.04). The number of subjects who received a diagnosis of TD in at least one of the 4 exams was 10 (40% period prevalence) in both treated groups. Thus, these results suggest that TD prevalence was greater in subjects treated with either drug than in untreated subjects of the same age, gender, and mental-retardation severity. Furthermore, there did not appear to be much difference in TD symptoms between subjects treated with these two drugs.
Combining results from Ganzini et al. and Sewell et al. – meta-analysis: Since the sample sizes of the first two studies reviewed above are relatively small, the MET-TD association was estimated somewhat imprecisely; i.e., the confidence intervals are wide, and it is difficult to rule out that each result is a chance finding. (It should be noted, however, that one cannot conclude that there is no effect or association simply because the p value is greater than 0.05 in each study; see Misinterpretation of P Values below.) Because the methods, study populations, and results are similar in these two investigations, we can combine the results using standard meta-analytic techniques. Results from the study of Matson et al. (2002) will not be included in this meta-analysis because the study population, diagnostic methods, and statistical methods differ from the other two studies.
There are two general statistical methods for conducting a meta-analysis: treating the exposure effects as fixed, i.e., assuming the same true effect in all studies; or treating unexplained variation across studies as random effects. A test for homogeneity of the OR across the two studies yielded a p value of 0.487, indicating that these results are compatible with the true OR being the same in both studies, i.e., suggesting a fixed-effects model (Rothman & Greenland, 1998:667-668). Therefore, I used the method of inverse-variance (precision) weighting to combine the results (Rothman & Greenland, 1998:660-661). The overall or combined OR was 2.24 (95% CI = 1.07, 4.70; p = 0.033). Thus, I estimate that the TD incidence rate would be more than twice as great among VA patients treated with MET for more than one month than if the same patients were not treated with MET.
Using the estimated prevalence ratio in each study, instead of the odds ratio, I estimated the overall or combined PR to be 1.81 (95% CI = 1.09, 3.01;p = 0.022). Note that although the PR estimate is closer to 1 than is the odds ratio (2.24), the 95% confidence interval around the PR estimate is narrower, and the two-sided p value is smaller. The reason for this improvement in precision and power is that the PR estimate from the Ganzini et al. paper is based on a matched analysis (of pairs), whereas the OR estimate in that study is based on an unmatched (crude) analysis.
It is important to interpret the above findings in light of several potential methodologic problems that limit our ability to make a causal inference about the effect of MET use on TD risk and to generalize our results to different populations.
Cross-sectional design: All three studies are effectively cross-sectional, not cohort (follow-up) studies, i.e., they involve data on TD prevalence, not incidence. There are three methodologic implications of this design:
a) Temporal ambiguity. It is possible that exposure to MET did not occur in some exposed cases until after TD developed, implying that MET could not be a cause of TD in those cases. It is unlikely, however, that this problem (often called “temporal ambiguity” or “temporality”) would account for the increased TD prevalence observed in MET users if there was no true effect of MET on TD risk. For temporal ambiguity to be a major source of bias in the MET-TD studies, TD status would have had to affect the prescription or use of MET in those subjects. If TD did, in fact, influence MET use, it was probably due to physicians discontinuing MET when they noticed TD symptoms in their patients. Whether or not this discontinuation resulted in remission of TD, the net result would be to underestimate the association between MET use and TD in a cross-sectional study – either because persistent TD cases would be classified as unexposed after discontinuation or because the duration of TD would be shorter among remitted cases with a history of exposure than among cases with no history of exposure. Thus, to the extent that temporality was a problem in the studies of Ganzini et al. and Sewell et al., it probably resulted in an underestimate of the true effect.
b) Selection bias. There may be selection bias if the selection of subjects is influenced by TD status and this influence differs for MET users and nonusers. This possibility is not likely to be an important source of bias in these studies because TD status was not known by the investigators – and probably not even by most subjects – when subjects were selected, and I doubt that TD status differentially influenced the use of medical care at the VA medical centers or the developmental center.
c) Risk vs. prognosis. Exposure to MET may have influenced the course of TD among cases, not necessarily its first occurrence in the population at risk – i.e., MET use may have exacerbated the severity or duration of the condition. If this scenario is true, we would say that MET exposure is a prognostic factor for TD outcome, but not necessarily a risk factor for TD. Although this possibility cannot be ruled out, it implies that MET use does have an adverse effect on the course of TD among cases, which might make the condition more persistent even after MET is discontinued. Alternatively, MET use might reduce mortality in TD cases, which would also account for the positive association observed between MET use and TD prevalence, but this possibility seems unlikely.
Confounding: Given the observational design, the results could have been confounded (biased) by unmeasured risk factors for TD. Although this possibility cannot be ruled out, it should be noted that the results from the studies Ganzini et al. and Sewell et al. are similar, and the investigators did control or check for confounding due to several TD risk factors. Furthermore, if there was residual confounding, I have no reason to suspect that the direction of the bias was more likely to be positive (exaggerating the true effect) than negative (underestimating the true effect). Suppose, for example, that race is a risk factor for TD, as suggested by the results of Morgenstern and Glazer (1993). The small difference in the proportion of whites between MET users and nonusers in the study of Sewell et al. (see Table 1) would not be nearly enough to explain the association observed between MET use and TD.
Misclassification bias: Misclassification of TD status or MET exposure might have led to bias in estimating the effect of MET use on TD. Nevertheless, the amount of such misclassification for each variable was most likely similar between categories of the other variable (i.e., nondifferential misclassification), in part because TD examiners were blinded to the MET treatment status of subjects. Consequently, the direction of the bias would probably have been toward the null, i.e., resulting in an underestimate of effect.
Lack of dose-response association: In none of the MET-TD studies did the authors report a “dose-response” (monotonic) association between dose or duration of MET use and TD prevalence. In fact, Ganzini et al. note that they did not find dose-response associations when such analyses were conducted (p. 1473). Although the lack of an observed dose-response relation might, in general, reflect the absence of a true effect, this evidence cannot be used to negate a causal interpretation, especially in these two studies. It is important to recognize that the investigators of most cross-sectional studies did not find dose-response associations with conventional antipsychotics, which are now widely regarded to be major risk factors for TD (Morgenstern & Glazer, 1993).
Unexpectedly high TD prevalence in nonusers of MET: Critics of these studies might argue that the results are inconsistent with other clinical studies of TD because the observed prevalence of TD in unexposed patients (often called “spontaneous dyskinesia”) was high: 18% in Ganzini et al. and 12% in Sewell et al. While these estimates might at first seem unexpected, there are several factors that could explain the high prevalence: the older age distribution of subjects; the high prevalence of other medical and psychiatric conditions such as organic brain disease, diabetes, and substance abuse that may increase the risk of TD; the high sensitivity for detecting relatively mild cases of TD; and the undetected previous use of MET and other neuroleptics by subjects in the “unexposed” groups (which could imply that the associations reported above are underestimates). Aside from this latter explanation, none of these factors necessarily implies that the estimated association between MET use and TD was biased.
Lack of generalizability: Some researchers or clinicians might argue that we cannot generalize the estimated MET-TD association obtained from the two studies of veterans to other adult populations in which MET might be prescribed. For such lack of generalizability to hold, however, there must exist certain TD risk factors that strongly modify the effect of MET use on TD risk, and the distribution of these effect modifiers would have to differ appreciably between veteran and non-veteran populations. In fact, from my knowledge of the TD literature, I know of no risk factors that would satisfy these conditions (often called “interactions”) for either MET or antipsychotics To place this issue in context, consider the well-known VA study in which the investigators found that coronary bypass surgery increased survival in patients with stable angina and three-vesse coronary artery disease (VA Coronary Artery Bypass Surgery Cooperative Group, 1984). Despite restriction of this study population to veterans, the resuls helped shape treatment decisions in non-veteran populations as well. Furthermore, the findings of Matson et al. (2002) from their study of adults with mental retardation are consistent with the results of the two veteran studies. Thus, although the findings presented in this document should be replicated in other populations, I have no reason to expect the results to be substantialy different.
Two additional points should be noted in regards to the issue of generalizability. First, lack of generalizability does not represent a type of bias in estimating exposure effects. These are different issues with very different implications to the interpretation of results. Second, in my estimation of TD risk among MET users (see Indirect Method above), I did not generalize TD risk from veterans to the general population. Rather, I extrapolated TD risk from regular users of conventional antipsychotics among outpatients included in other studies of TD incidence. Of course, we cannot be sure how well the background TD risk in this psychiatric population (in the absence of antipsychotic use) reflects the background risk in the general population of MET users, but this issue cannot be addressed without conducting large comparative studies of TD incidence in multiple populations of MET users and nonusers.
Misinterpretation of P Values
Some researchers and practitioners would claim that the results of Ganzini et al. and Sewall et al. are “not statistically significant” – i.e., the two-sided p value for testing the null hypothesis of no association is greater than 0.05 or, equivalently, the 95% confidence interval around the estimated odds ratio includes the null value of one. Despite the widespread practice of labeling studies as “significant” or “non-significant,” this practice has no scientific merit and is now strongly discouraged by quantitatively sophisticated researchers, including epidemiologists and statisticians. More importantly, interpreting study results on the basis of whether p < or> 0.05 is often very misleading. Indeed, inferring from the results of Ganzini et al. or Sewall et al. that there is no effect of MET use on the occurrence of TD because the p value is greater than 0.05 is simply incorrect. This view is consistent with numerous publications in the scientific literature in the past two decades, including textbooks by Anders (1993), Rothman and Greenland (1998), Szklo and Nieto (2000), Rothman (2002), Aschengrau and Seage (2003), Savitz (2003), and Jewell (2004).
Interpretation of statistical findings should focus on the magnitude of the association and the precision with which the association is estimated (e.g., the estimated odds ratio and 95% confidence interval), rather than values – or worse, whether p < or > 0.05. The problem with p values is that they depend on both the magnitude of the association as well as the number of subjects in the study. Over-reliance on p values and significance testing is probably due, at least partly, to a misunderstanding of p values. It is important to appreciate, for example, that ap value is not the probability that the null hypothesis is true or that the result was due to chance. Rather, the p value is the probability of obtaining the observed result (technically, test statistic) or a more extreme value if the null hypothesis (no association) were true.
Misinterpretation of a Relative Risk < 2
Some researchers and legal experts would argue that the plaintiff's condition, tardive dyskinesia, is “more likely than not” caused by factors other than MET whenever the best estimate of the relative risk (RR) – i.e, the TD risk in persons exposed to MET divided by the TD risk in comparable unexposed persons – is less than 2. The rationale for this position is based on another epidemiologic measure called the attributable fraction in the exposed population (AFe), which is the proportion of exposed cases (new TD cases who used MET) that was attributable to MET use – i.e., the probability that a randomly selected exposed case (like the plaintiff) would not have developed TD during a given period in the absence of exposure. Since the AFe is equal to (RR-1)/RR, an RR less than 2 implies that the AFe is less than 50%. Although I did not actually estimate the RR in my analyses, it would be a little less than the estimated OR of 2.24 reported earlier, perhaps closer to the PR estimate of 1.81. The difference between the RR and OR depends on the duration of MET exposure (and follow-up) – the longer the exposure, the larger the difference between these two measures.
The argument summarized above is fallacious and potentially very misleading: An AFe less than 50% does not mean that the plaintiff's TD was more likely than not caused by factors other than MET exposure. The fundamental problem with the argument is that it confuses two different concepts: the probability that the plaintiff's exposure to MET was a contributory cause of his or her TD (called the probability of causation, PC); and the proportion of exposed cases occurring during a given period that would not have occurred in the absence of exposure (the AFe). While we can estimate the AFe readily from epidemiologic studies, the parameter of interest to address the legal criterion for causation (“more likely than not”) is the PC, not the AFe. The proportion of exposed cases attributable to the exposure is not the same as the proportion of exposed cases caused by the exposure. More importantly, the AFe can be much less, but not greater than the PC because the numerator of the PC includes any exposed case that would have occurred at a later time or would not have occurred at all in the absence of exposure, whereas the numerator of the AFe includes only those exposed cases that would not have occurred at all in the absence of exposure. Although the AFe can be estimated from epidemiologic data, the PC cannot be estimated from epidemiologic data without certain restrictive assumptions involving unknown biological mechanisms. Thus, an estimated RR less than 2, even if it is valid, does not imply that the PC is less than 50%.
Confusion between the PC and AFe and misleading interpretations of AFe estimates are common in epidemiology, biostatistics, and the law. Nonetheless, the points raised above have been clearly and thoroughly discussed in the peer-reviewed literature for two decades (e.g., Greenland & Robins, 1988; 2000; Robins & Greenland, 1989; Beyea & Greenland, 1999; Greenland, 1999). Despite these discussions, the Reference Manual on Scientific Evidence, which is published by the Federal Judicial Center and often cited by attorneys, fails to distinguish between the PC and AFe (called the “attributable risk” in that document; Green et al., 2000:381- 386). Although those authors refer to the paper by Greenland and Robins (1988) in footnote 142 (p. 385), they do not mention the implication of the relative risk being less than 2. Furthermore, that manual does not distinguish between the incidence rate and risk of disease – concepts that are crucial for understanding other epidemiologic measures (Green et al., 2000:348-349; see also Measures of Disease Frequency above).
TD Risk Among Users of MET for More Than 12 Weeks
The second part of my testimony deals with the risk of TD among long-term users of MET. The general warning in the package insert for MET states that “Extrapyramidal symptoms, manifested primarily as acute dystonic reactions, occur in approximately 1 in 500 patients treated with the usual adult dosages of 30-40 mg/day of metocloproamide.” Under the warning for “Tardive Dyskinesia,” no risk or prevalence estimates are provided, but the claim is made that “it is impossible to predict which patients are likely to develop the syndrome.” Not only is this claim inconsistent with vast TD literature, but the quantitative implication of the warning is unclear. It seems to imply that the risk of TD, another category of “extrapyramidal symptoms,” is extremely low – purportedly less than 1 in 500.
To assess the accuracy of these statements in the package insert, we can first take a qualitative approach by considering their logical consistency with published results of TD prevalence among MET users in the three cross-sectional studies reviewed above. The estimated (point) prevalence of TD among users of MET for more than 12 weeks was 29% in the Ganzini et al. study and 27% in the Sewell et al. study. The 1-year period prevalence among users of MET for at least 4 years was 40% in the Matson et al. study.
Suppose that 1,000 adults initially without TD are followed for several years from the time they start using MET, and they are examined frequently – say every three months – for the detection of TD using standard diagnostic procedures and criteria (Schooler & Kane, 1982). If the true risk of TD in this population were no greater than 1 in 500 (0.002), as implied by the MET package insert, we would expect no more than two incident cases to occur during the course of treatment. Now suppose that a cross-sectional study is conducted in this population several years after the start of MET use. Regardless of how long symptoms of TD persist in the two incident cases and regardless of who might be lost to follow-up, we would not expect to observe more than two prevalent cases at any given time or during any 1-year interval. Thus, it would be mathematically impossible for the prevalence of TD to be greater than 25% in that population (see Measures of Disease Frequency – Point prevalence). Consequently, there is no way that a TD risk of only 0.002 could be consistent with the three estimates of TD prevalence obtained from the studies cited above unless the investigators of all three studies selected MET users who were known to have TD before their inclusion in the studies. Such a scenario would be a serious violation of scientific as well as ethical principles, and there is no evidence that such violations occurred.
The above reasoning, which involves simple epidemiologic principles and logic without any new methods or questionable assumptions, suggests that the risk of TD cannot be 1 in 500 or less. To check for empirical evidence, I searched the biomedical literature for studies in which the investigators estimated the incidence rate of TD in a population of MET users (see below).
Estimating the Incidence Rate of TD Among MET Users: Direct Method
To the best of my knowledge, only one attempt has been made to estimate the incidence rate of TD among users of MET. Using data on MET sales and neurological complications reported by physicians to the Swedish Adverse Drug Reactions Advisory Committee between 1977 and 1981, Wiholm et al. (1984) estimated the incidence rate of TD among MET users in Sweden to be 36-50 per 100,000/year. Although these estimates are much less than the incidence rates reported for TD in epidemiologic studies of psychiatric patients treated with antipsychotics (in the order of 5,000 per 100,000/year = 0.05/year), I believe the rates reported by Wiholm et al. are dramatically underestimated – perhaps by 100-fold or more. There are five reasons for this assessment.
First, Wiholm et al. did not measure TD symptoms in any patients, but relied on a passive surveillance system to identify cases. Second, in 1977-81, physicians who were prescribing MET were unfamiliar with TD symptoms or their possible link with MET; thus, they would have been unlikely to identify and report dyskinetic movements in their patients. To put this in perspective, consider Yale psychiatrists who were trained in the same period to identify TD at the Connecticut Mental Health Center. In our early epidemiologic studies of outpatients in that facility, we found that most TD cases had not been reported on the medical records or referred to the TD Clinic by those Yale psychiatrists. Third, even when physicians are taught how to identify TD symptoms, it is doubtful that most mild or moderate cases will be identified and reported; yet most TD cases in epidemiologic studies are mild or moderate in severity. Fourth, there may be a lag of many weeks, months, or even years between first use of a neuroleptic and the occurrence of TD. Thus, the treating physician may not make the connection between drug use and TD, especially if they have not been alerted to its symptoms. Fifth, Wiholm et al. were not able to determine from their national data the distribution of duration of MET use in the Swedish population. Thus, they could not accurately assess how the TD rate varied by duration of MET use.
In summary, therefore, I do not believe the results of Wiholm et al. (1984) tell us anything about the true incidence rate of TD among long-term MET users. Indeed, even Wiholm et al. concluded in their abstract that “Long term treatment with metoclopramide is accompanied by a substantial risk of developing tardive dyskinesia especially among elderly people.” The best available evidence for estimating TD risk, I believe, comes from extrapolating the results of other TD studies (see Indirect Method below).
Estimating the Incidence Rate and Risk of TD in MET Users: Indirect Method
Despite the critical limitations of the study by Wiholm et al. (1984) for estimating TD incidence (see above), it is possible to estimate the incidence rate and risk of TD among MET users indirectly by combining results from three sources: 1) my meta-analysis of the results from Ganzini et al. and Sewell et al. (see above); 2) my published meta-analysis of 21 studies examining the association between antipsychotic use and TD (Morgenstern et al., 1987); and 3) results from published studies of TD incidence during the first several years of exposure to conventional antipsychotic medications.
In my meta-analysis of the results from Ganzini et al. and Sewell et al., I estimated that the combined incidence rate ratio, comparing MET users with nonusers, was 2.24 (95% CI = 1.07, 4.70; two-sided p value = 0.033). In my meta-analysis of 21 published studies (Morgenstern et al., 1987), I estimated that the incidence rate ratio, comparing psychiatric outpatients treated with conventional antipsychotics with psychiatric outpatients not treated with antipsychotics, was 2.87 (95% CI = 2.34, 3.52). Comparing this latter result to the former, it appears that the effect on TD occurrence of MET in commonly prescribed doses is 2.24/2.87 = 78% the effect of conventional antipsychotics in commonly prescribed doses before 1986.
We can estimate the risk (cumulative incidence) of TD in MET users, without conducting an incidence study of MET users, by extrapolating from the incidence findings of other studies of conventional antipsychotics. There are 4 published studies in which the investigators provided data to estimate the average incidence rate of TD during the first several years of antipsychotic use. These studies and estimated incidence rates are summarized in the table below.
* A presumptive diagnosis of TD is based on the subject meeting minimum symptom criteria or one exam; a persistent diagnosis of TD is based on the subject meeting those criteria on two consecutive exams.
† The average rate was derived from the published risk by assuming that the rate remained constant during the follow-up period (see Measures of Disease Frequency – Risk).
As expected, the average incidence rate is strongly related to the mean age of subjects: the rate is 4-5 times greater in elderly populations than in young-adult populations. To estimate TD risk during the first 5 years of MET exposure, I extrapolated from the results of three studies: 1) the study of young adults with the lowest incidence rate (Kane et al., 1984); 2) the study of predominantly middle-aged adults (Glazer et al., 1993); and 3) the study of elderly adults with the highest incidence rate (Woerner et al., 1998). By multiplying the rate in each of those studies by 2.24/2.87 = 0.78, we obtain incidence rates among MET users of 0.0306 per year, 0.0618 per year, and 0.1595 per year, respectively. Assuming these rates are constant over time, we can use the estimated incidence rates to derive the risks of TD (see Measures of Disease Frequency – Risk). These estimated risks of TD (in %), by total years of MET use, are shown in the table below (rounded to the nearest percent).
a. Derived from Kane et al. (1984).
b. Derived from Glazer et al. (1993).
c. Derived from Woerner et al (1998).
From these results, we see that the 5-year risk of TD after 5 years of using that drug is 14% (1 in 7) for young adults, 27% (1 in 4) for middle-aged adults, and 55% (1 in 2) for elderly adults. Similarly, the TD risk after one year of MET use is 3% (1 in 33) for young adults, 6% (1 in 17) for middle-aged adults, and 15% (1 in 7) for elderly adults. Thus, even assuming the lowest TD rate reported in the literature for outpatients maintained on antipyschotics and a correction factor (78%) to account for a lower rate among MET users, these estimates of TD risk among patients treated with MET are much greater than 1 in 500 (0.2%), the maximum risk implied in the package insert for metoclopramide/Reglan. Furthermore, this interpretation is consistent with Pasricha et al. (2006), who concluded that TD risk is “grossly underestimated in the package insert.”
Summary and Conclusion
Despite the methodologic limitations of the three epidemiologic studies by Ganzini et al. (1993), Sewell et al. (1994), and Matson et al. (2002), their results – along with relevant findings from other investigations – lead me to conclude with a reasonable degree of medical certainty that use of metoclopramide for more than 12 weeks increases the risk of tardive dyskinesia. This conclusion is consistent with those expressed in two review papers by Jimenez-Jimenez et al (1997) and Lata & Pigarelli (2003). Although the results of these three cross-sectional studies may be biased, I believe their findings are more likely to reflect an underestimate than an overestimate of the true effect of metoclopramide on TD occurrence. Furthermore, there is no evidence from any studies of TD to suggest that the results of Ganzini et al. and Sewall et al. cannot be generalized to non-veterans in the general population, a conclusion that is supported by the consistent results found by Matson et al. for a different non-veteran population.
Finally, although no incidence studies of TD have been conducted among users of metoclopramide, I showed that a TD risk of 1 in 500 or less is logically inconsistent with prevalence findings obtained by Ganzini et al, Sewell et al., and Matson et al In addition, I was able to use the results of Ganzini et al. and Sewell et al., along with other studies of TD, to estimate the risk of TD indirectly among persons treated with metoclopramide for various durations. My results suggest that the magnitude of the metoclopramide effect is not much smaller than the effect of conventional antipsychotic use on TD risk. Furthermore, the best estimate of TD risk among users of metoclopramide is much greater than the risk implied by the warning statements in the package insert.
Anders A. Biostatistics for Epidemiologists. Boca Raton: Lewis Publishers, 1993, pp. 40-42.
Aschengrau A, Seage GR III. Essentials of Epidemiology in Public Health. Sudbury, MA: Jones and Bartlett Publishers, 2003, pp. 307-310.
Beyea J, Greenland S. The importance of specifying the underlying biologic model in estimating the probability of causation. Health Physics 1999; 76:269-274.
Chakos MH, Alvir JMJ, Woerner MG, et al. Incidence and correlates of tardive dyskinesia in first episode of schizophrenia. Archives of General Psychiatry 1996; 53:313-319.
Ganzini L, Casey DE, Hoffman WF, Heintz RT. Tardive dyskinesia and diabetes mellitus. Psychopharmacology Bulletin 1992; 28:281-286.
Ganzini L, Casey DE, Hoffman WF, McCall AL The prevalence of metoclopramide-induced tardive dyskinesia and acute extrapyramidal movement disorders. Archives of Internal Medicine 1993; 153:1469-75.
Ganzini L, Heintz RT, Hoffman WF, Casey DE. The prevalence of tardive dyskinesia in neuroleptic-treated diabetics: A controlled study. Arch Gen Psychiatry 1991; 48:259-263.
Glazer WM, Morgenstern H, Doucette JT. Predicting the long-term risk of tardive dyskinesia in outpatients maintained on neuroleptic medications. Journal of Clinical Psychiatry 1993; 54:133- 139.
Green MD, Freedman DM, Gordis L. Reference guide on epidemiology. In: Reference Manual on Scientific Evidence, 2nd ed. Federal Judicial Center, 2000:333-400 (see pp. 348-349, 351- 352, 381-386). Available at http:// www.fjc.gov/public/pdf.nsf/lookup/scimanO0.pdf.
Greenland S. The relation of the probability of causation to the relative risk and the doubling dose: A methodologic problem that has become a social problem. Am J Public Health 1999; 89:1166-1169.
Greenland S, Robins JM. Conceptual problems in the definition and interpretation of attributable fractions. Am J Epidemiol 1988; 128:1185-1197.
Greenland S. Robins JM. Epidemiology, justice, and the probability of causation. Jurimetrics 2000; 40:321-340.
Grimes JD, Hassan MN, Preston DN. Adverse neurologic effects of metoclopramide. Canadian Medical Association Journal 1982; 126:23-25.
Jeste DV, Wyatt RJ Understanding and Treating Tardive Dyskinesia. New York: Guilford Press, 1982, pp. 188, 209.
Jewell NP. Statistics for Epidemiology. Baco Raton: Chapman & Hall/CRC, 2004, pp. 61-62.
Jimenez-Jimenez FJ, Garcia PJ, Molina JA. Drug-induced movement disorders. Pharmacoepidemiology 1997; 16:180-204.
Kane JM, Woemer M, Weinhold P, Wegner J, Kinon B. Incidence of tardive dyskinesia: Five- year data from a prospective study. Psychopharmacology Bulletin 1984; 20:39-40.
Lata PF, Pigarelli DLW. Chronic metoclopramide therapy for diabetic gastroparesis. Ann Pharmacotherpy 2003; 37:122-126.
Matson JL, Mayville EA, Bielecki J, Smalls Y, Eckholdt CS. Tardive dyskinesia associated with metoclopramide in persons with developmental disabilities. Research in Developmental Disabilities 2002; 23:224-233.
Morgenstem H, Glazer WM. Identifying risk factors for tardive dyskinesia among long-term outpatients maintained with neuroleptic medications. Archives of General Psychiatry 1993; 50:723-733.
Morgenstern H, Glazer WM, Niedzwiecki D, Nourjah P. The impact of neuroleptic medication on tardive dyskinesia: A meta-analysis of published studies. American Journal of Public Health 1987; 77:717-724.
Pasricha PJ, Pehlivanov N, Sugumar A, Jankovic J. Drug insight: From disturbed motility to disordered movement-a review of the clinical benefits and medicolegal risks of metoclopramide. Nature Clinical Practice Gastroenterology & Hepatology 2006; 3:138-148.
Robins JM, Greenland S. Estimability and estimation of excess and etiologic fractions. Stat Med 1989; 8:845-859.
Rothman KJ. Epidemiology: An Introduction. New York: Oxford University Press, 2002, pp. 114-123.
Rothman KJ, Greenland S. Modern Epidemiology, 2nd ed. Philadelphia: Lippincott-Raven Publishers, 1998, pp. 64,186-188, 660-661, 667-668.
Savitz DA. Interpreting Epidemiologic Evidence: Strategies for Study Design and Analysis. New York: Oxford University Press, 2003, pp. 248-251.
Schooler NR, Kane JM. Research diagnoses for tardive dyskinesia (letter). Archives of General Psychiatry 1982; 39:486-487.
Sewell DD, Jeste DV. Metoclopramide-associated tardive dyskinesia: An analysis of 67 cases. Archives of Family Medicine 1992; 1:271-278.
Sewell DD, Kodsi AB, Caligiuri MP, Jeste JV. Metoclopramide and tardive dyskinesia. Biological Psychiatry 1994; 36:630-32.
Shaffer D, Butterfield M, Pamer C, Mackey AC. Tardive dyskinesia risks and metoclopramide use before and after U.S. market withdrawal of cisapride. Journal of the American Pharmacists Association 2004; 44:661-665.
Stewart RB, Cerda JJ, Moore MT, Hale WE. Metoclopramide: An analysis of inappropriate long-term use in the elderly. Annals of Pharmacotherapy 1992; 26:977-979.
Szklo M, Nieto FJ. Epidemiology: Beyond the Basics. Gaithersburg, MD: Aspen Publishers, 2000, pp. 416-419.
Veterans Administration Coronary Artery Bypass Surgery Cooperative Study Group. Eleven- year survival in the Veterans Administration Randomized Trial of coronary bypass surgery for stable angina. New England Journal of Medicine 1984; 311:1333-1339.
Wiholm B-E, Mortimer O, Goethius G, Haggstrom JE. Tardive dyskinesia associated with metoclopramide. British Medical Journal 1984; 288:545-547.
Woerner MG, Alvir JMJ, Saltz BL, Lieberman JA, Kane JM. Prospective study of tardive dyskinesia in the elderly: Rates and risk factors. American Journal of Psychiatry 1998; 155:1521-1528.