In re Abilify (Aripiprazole) Prods. Liab. Litig. (299 F. Supp. 3d 1291, 2018)

M. CASEY RODGERS, CHIEF UNITED STATES DISTRICT JUDGE

This is a multidistrict product liability action against the manufacturers and marketers of the prescription drug Aripiprazole, more commonly known as Abilify. Plaintiffs allege that, after taking Abilify as prescribed, they developed impulsive and irrepressible urges to engage in certain harmful behaviors, including impulsive gambling, eating, shopping, and sex. Defendants deny the allegations and maintain that Abilify could not, and did not, cause Plaintiffs' impulse control problems.

Defendants have moved for summary judgment on the issue of general causation-that is, whether Abilify is capable of causing uncontrollable impulses to engage in certain harmful behaviors. See ECF No. 428. Both the motion, see id. , and the response, see ECF No. 463, are supported by expert testimony. Each side challenges the other's experts as unreliable and those motions are also pending. A four-day evidentiary hearing was conducted jointly with Magistrate Judge Gary R. Jones of this Court, and Judge James J. Deluca of the New Jersey Superior Court, who presides over multiple similar cases in New Jersey state court. Now, having carefully considered the law, the voluminous record, and the parties' arguments, the Court concludes that Plaintiffs have satisfied their burden to demonstrate that a genuine dispute of material fact exists as to whether Abilify can cause uncontrollable impulsive behaviors in individuals taking the drug.

I. Background

Abilify is an atypical antipsychotic drug developed and manufactured by Defendants Otsuka Pharmaceutical Co., Ltd. and Otsuka America Pharmaceutical, Inc., who jointly market and distribute it in the United States with Defendant Bristol-Myers Squibb Company (collectively, "Defendants"). See Master Complaint, ECF No. 108-1 at 5. In 2002, Abilify was approved by the Food and Drug Administration ("FDA") for the treatment of schizophrenia. Since then, Abilify also has been approved for use in patients with bipolar disorder, irritability associated with autistic disorder, Tourette's Syndrome, and as an add-on treatment for major depressive disorder. See Product Label, ECF No. 428-1 at 2. "[T]ens of millions of patients worldwide have used Abilify to help manage the symptoms of these very debilitating mental health conditions." See DSJ, ECF No. 428-26 at 9.

In 2010, the first published reports suggesting a possible link between Abilify and pathological gambling began appearing in the medical literature. More published reports followed, as well as hundreds of informal reports from patients and healthcare professionals to Defendants and the FDA, describing the onset of impulsive gambling and other impulse control disorders in patients treated with Abilify. The scientific community, the FDA, Defendants, and public health agencies worldwide took notice and began examining whether Abilify is linked to impulse control disorders. The research findings and conclusions of these bodies are at the heart of the motions currently pending before this Court.

In 2012, following a safety review of Abilify based on reports of pathological gambling with patients' use of the drug, the European Medicines Agency ("EMA") required Defendants to modify the drug's product label in Europe to include pathological gambling as a possible "post-marketing undesirable effect" of Abilify and to warn of an "increased risk" of pathological gambling in Abilify patients with a prior history of gambling. See FDA Pharm. Vigil., ECF No. 428-11 at 5, 12. In November 2015, Health Canada also found an "increased risk" of pathological gambling, as well as hypersexuality, with Abilify use and ordered that the drug's product monograph in Canada be updated to advise of these possible adverse effects. See id. at 5, 12. Health Canada's safety review and subsequent product monograph update prompted the FDA to initiate a pharmacovigilance review in the United States to evaluate whether the potential link between Abilify and impulse control disorders presented a "safety issue warrant[ing] any regulatory action." See id. at 5. The FDA's review identified an association between Abilify and impulse control disorders, based on an analysis of cases in its adverse event reporting database (FAERS), the published scientific literature, and Defendant's clinical trial and post-marketing patient data. See id. at 4, 29. On May 3, 2016, the FDA issued a safety warning that "uncontrollable and excessive urges" to "gamble, binge eat, shop and have sex" had been reported with the use of Abilify, even in patients with no prior history of impulsive behaviors. In August 2016, the FDA required Defendants to modify Abilify's product label in the United States to warn of "post-marketing case reports suggest[ing] that patients can experience intense urges, particularly for gambling, and the inability to control these urges while taking" the drug. See Product Label, ECF No. 428-1 at 2, 24. The United States product label was also modified to warn of "[o]ther compulsive urges, reported less frequently, [which] include: sexual urges, shopping, eating or binge eating, and other impulsive or compulsive behaviors." See id. At that point, Abilify had been on the market in the United States for almost 14 years.

A short biochemistry discussion may be helpful at this point. The human brain is a tremendously complex biochemical system. It contains billions of interconnected nerve cells, called neurons, that use chemical and electrical signals to send information throughout the body. The function of a neuron is to process and transmit information-it receives signals from other neurons, integrates and interprets those signals, and transmits signals to other, adjacent neurons. The signals within neurons are carried throughout the brain in the form of electrical impulses. When a signal is sent from one neuron to another, it must cross a microscopic gap between the two communicating neurons. This gap is called a synapse or synaptic cleft. At the synapse, the electrical signal within the neuron is converted to a chemical signal and sent across the synapse towards the receiving neuron. This chemical signal is transported by molecules, called neurotransmitters, that attach to special structures on the outer surface of the receiving neuron, called receptors. There are many different types of receptors, categorized by the type of neurotransmitters with which they interact. The attachment of neurotransmitters to receptors can either stimulate or inhibit electrical activity in the receiving neuron, depending on which neurotransmitter is released and which receptors it activates. In any one synapse, there may be hundreds of neurotransmitters continually moving between, and acting on, neurons, triggering varying physiological effects throughout the brain and the body. Any disruption to the neuronal communication process-whether to the production, release, or attachment of the various neurotransmitters-can alter brain function and, as it relates to this case, human behavior.

Dopamine is a neurotransmitter in the central nervous system that is believed to play an integral role in a number of physiological processes, including movement, cognition, emotional stability, and, relevant to this case, reward-motivated behaviors. It acts on five different receptors-D1, D2, D3, D4, and D5-along four major pathways in the brain-the nigrostriatal pathway, the mesocortical pathway, the mesolimbic pathway, the tuberoinfundibular pathway. This case is primarily concerned with the activity of dopamine in the mesolimbic pathway, which regulates pleasure, reward processing, and motivation. Under normal circumstances, the brain responds to rewarding activities or stimuli by releasing dopamine into the mesolimbic pathway, where it binds with dopamine receptors to produce feelings of pleasure. As dopamine levels subside, so do the feelings of pleasure. If the rewarding activity is repeated, then dopamine is again released, and more feelings of pleasure are produced. The release of dopamine and the resulting pleasurable feelings serve as positive reinforcements that motivate repetition of the pleasure-inducing activity.

Pharmaceutical companies create drugs that can mimic, duplicate, or block the activity of natural, or "endogenous," dopamine in the brain. The effect of a given drug depends on two pharmacological properties that relate to the manner in which the drug interacts with dopamine receptors: affinity and intrinsic activity. Affinity refers to whether and how tightly the drug binds to dopamine receptors. Intrinsic activity refers to the degree to which the drug, once bound, activates dopamine receptors to produce a measurable physiological effect. Based on these properties, drugs that bind to dopamine receptors can act as agonists or antagonists. A full agonist has both high affinity and 100% intrinsic activity, meaning that it binds tightly to dopamine receptors and mimics the activity of dopamine, producing the same level of physiological response that dopamine naturally produces. Antagonists bind to dopamine receptors, but produce no physiological effects; instead, they simply occupy a receptor site, thereby preventing endogenous dopamine from binding to and activating it. A partial agonist binds to dopamine receptors, but produces less of a response than a full agonist. The functional activity of some partial agonists depends on the presence or absence of endogenous dopamine in the surrounding area. Where dopamine concentrations are high, the partial agonist functions as an antagonist (i.e. , functional antagonist), but where dopamine concentration is low, the partial agonist functions as a full agonist (i.e. , functional agonist). In this case, Plaintiffs' position as to how Abilify causes impulse control problems centers on how the drug binds and interacts with two dopamine receptors-D2and D3-to produce physiological effects in the form of impulsive behaviors.

II. Expert Challenges

To establish general causation, Plaintiffs have proffered the testimony of five experts: Dr. Antoine Bechara, Dr. Joseph Glenmullen, Dr. Eric Hollander, Dr. Russell V. Luepker, and Dr. David Madigan. Simply stated, each of Plaintiffs' experts opines that Abilify can cause impulsive behaviors and each presents scientific evidence in support of his conclusion. Defendants challenge the admissibility of Plaintiffs' expert testimony on general causation as unreliable under Federal Rule of Evidence 702 and Daubert v. Merrell Dow Pharmaceuticals, Inc. , 509 U.S. 579, 113 S.Ct. 2786, 125 L.Ed.2d 469 (1993). More specifically, Defendants contend that Plaintiffs have failed to provide reliable scientific evidence demonstrating a statistically significant association between Abilify and impulsive behaviors. According to Defendants, this omission is a fatal flaw in Plaintiffs' case because their remaining evidence is insufficient as a matter of law, even in the aggregate, to establish general causation by Eleventh Circuit standards. Defendants rely on five experts of their own-Dr. Pierre Blier, Dr. Deborah Leiderman, Dr. Marc Potenza, Dr. Douglas Weed, and Dr. Catherine Winstanley-to support their position that Abilify cannot cause impulse control disorders, each of whom submitted an opinion that Plaintiffs in turn challenge as unreliable.

A. Legal Standard for Expert Testimony

Rule 702, as explained by Daubert and its progeny, governs the admissibility of expert testimony. Rink v. Cheminova, Inc. , 400 F.3d 1286, 1291 (11th Cir. 2005). Under Rule 702 and Daubert , district courts are compelled to act as "gatekeepers" to ensure the reliability and relevancy of expert testimony. Id. (quoting Daubert , 509 U.S. at 589, 113 S.Ct. 2786 ). Expert testimony is reliable and relevant-and, therefore, admissible-when the following criteria are met: (1) the expert is sufficiently qualified to testify about the matters he intends to address; (2) the methodology used is "sufficiently reliable as determined by the sort of inquiry mandated in Daubert ; and (3) the testimony assists the trier of fact, through the application of scientific, technical, or specialized expertise, to understand the evidence or to determine a fact in issue." Id. The Eleventh Circuit refers to these criteria separately as "qualification, reliability, and helpfulness," United States v. Frazier , 387 F.3d 1244, 1260 (11th Cir. 2004), and has emphasized that they are "distinct concepts that courts and litigants must take care not to conflate," Quiet Tech. DC-8, Inc. v. Hurel-Dubois UK Ltd. , 326 F.3d 1333, 1341 (11th Cir. 2003). The party offering the expert has the burden of showing, by a preponderance of the evidence, that each of these requirements is met. Rink , 400 F.3d at 1292.

To meet the qualification requirement, a party must show that its expert has sufficient "knowledge, skill, experience, training, or education to form a reliable opinion about an issue that is before the court." Hendrix ex.Rel. G.P. v. Evenflo Co., Inc. , 609 F.3d 1183, 1193 (11th Cir. 2010) (citing Fed. R. Evid. 702 ) (" Hendrix II "), aff'g 255 F.R.D. 568 (N.D. Fla. 2009) (" Hendrix I "). The qualifications standard for expert testimony is "not stringent" and "[s]o long as the witness is minimally qualified, objections to the level of [his] expertise [go] to credibility and weight, not admissibility." Hendrix I , 255 F.R.D. at 585.

To meet the reliability requirement, an expert's opinion must be based on scientifically valid principles, reasoning, and methodology that are properly applied to the facts at issue. Frazier , 387 F.3d at 1261-62. The reliability analysis is guided by several factors, including: (1) whether the scientific technique can be or has been tested; (2) whether the theory or technique has been subjected to peer review or publication; (3) whether the technique has a known or knowable rate of error; and (4) whether the technique is generally accepted in the relevant community. Daubert , 509 U.S. at 593-94, 113 S.Ct. 2786. "[T]hese factors do not exhaust the universe of considerations that may bear on the reliability of a given expert opinion, and a federal court should consider any additional factors that may advance its Rule 702 analysis." Quiet Tech. , 326 F.3d at 1341. The court's focus must be on the expert's principles and methodology, not the conclusions they generate. Daubert , 509 U.S. at 595, 113 S.Ct. 2786. The test for reliability is "flexible" and courts have "broad latitude" in determining both how and whether this requirement is met. Kumho Tire Co., Ltd. v. Carmichael , 526 U.S. 137, 141-42, 119 S.Ct. 1167, 143 L.Ed.2d 238 (1999).

Finally, to satisfy the helpfulness requirement, expert testimony must be relevant to an issue in the case and offer insights "beyond the understanding and experience of the average citizen." United States v. Rouco , 765 F.2d 983, 995 (11th Cir. 1985). Relevant expert testimony "logically advances a material aspect of the proposing party's case" and "fits" the disputed facts. McDowell v. Brown , 392 F.3d 1283, 1298-99 (11th Cir. 2004). Expert testimony does not "fit" when there is "too great an analytical gap" between the facts and the proffered opinion. Gen. Elec. Co. v. Joiner , 522 U.S. 136, 147, 118 S.Ct. 512, 139 L.Ed.2d 508 (1997).

When scrutinizing the reliability and relevance of expert testimony, a court must remain mindful of the delicate balance between its role as a gatekeeper and the jury's role as the ultimate factfinder. Frazier , 387 F.3d at 1272. The court's gatekeeping role "is not intended to supplant the adversary system or the role of the jury." Allison v. McGhan Med. Corp. , 184 F.3d 1300, 1312 (11th Cir. 1999). Only the jury may determine "where the truth in any case lies" and the court "may not usurp this function." Frazier , 387 F.3d at 1272. Thus, a court may not "evaluate the credibility of opposing experts" or the persuasiveness of their conclusions, Quiet Tech. , 326 F.3d at 1341 ; instead, its duty is limited to "ensur[ing] that the fact-finder weighs only sound and reliable evidence, Frazier , 387 F.3d at 1272.

B. Reliability of Expert Testimony on General Causation

To prevail in a pharmaceutical products liability case, a plaintiff must establish both general and specific causation through reliable expert testimony. Chapman v. Procter & Gamble Distributing, LLC , 766 F.3d 1296, 1303-04 (11th Cir. 2014). General causation is established by demonstrating, often through a review of scientific or medical literature, that a drug or chemical can, in general, cause the type of harm alleged by the plaintiff. See Hendrix II , 609 F.3d at 1196. Specific causation is established by showing that exposure to the allegedly toxic drug or chemical actually caused an individual plaintiff's injury. Id. Only general causation-whether Abilify is capable of causing impulse control disorders, such as impulsive gambling-is at issue in the motions currently pending before the Court.

The Eleventh Circuit has developed an extensive body of Daubert jurisprudence around the reliability of different categories of scientific evidence that may support an expert opinion on general causation. The Eleventh Circuit considers three "primary" methodologies "indispensable" for proving that a drug can cause a specific adverse effect: epidemiological studies, dose-response relationship, and background risk of disease. Chapman , 766 F.3d at 1308. A general causation opinion that is not supported by at least one of these primary methodologies is unreliable as a matter of law. See id. An expert who has reliably applied primary methodologies may bolster his general causation opinion with evidence from "secondary" methodologies, such as: biological plausibility, case studies and adverse event reports, extrapolations from animal and in vitro studies, and extrapolations from analogous drugs. See id. Importantly, the flaws inherent in the secondary methodologies limit their reliability under Daubert. See id. For this reason, secondary methodologies alone, even in the aggregate, cannot establish general causation. See id. ; see also Hendrix II , 609 F.3d at 1197.

In this case, the parties' experts offer various combinations of primary and secondary methodologies in support of their general causation opinions. To frame the Court's analysis of the expert opinions, a brief review of the scientific and legal principles governing the reliability of each methodology follows.

1. Primary Methodologies

a. Epidemiological Studies

The "best evidence of causation in toxic tort actions" is grounded in epidemiology, Rider v. Sandoz Pharmaceuticals Corp. , 295 F.3d 1194, 1199 (11th Cir. 2002), which is the branch of science that studies the incidence, distribution, and cause of disease in human populations, Reference Manual on Scientific Evidence at 551 ("Ref. Man."). The first step in establishing causation through epidemiology is to demonstrate that exposure to a drug is associated with a particular disease or adverse effect. Ref. Man. at 566. Once an association is identified, scientists next determine whether the association represents "a true cause-effect relationship" between exposure and the disease. Ref. Man. at 597. This is the sine qua non of general causation.

This causation inquiry is guided by nine well-established factors, known in the scientific community as the Bradford Hill factors. These include: (1) temporal relationship; (2) strength of the association; (3) dose-response relationship; (4) replication of the findings; (5) biological plausibility; (6) consideration of alternative explanations; (7) cessation of exposure; (8) specificity of the association; and (9) consistency with other knowledge. See Ref. Man. at 599-600. No one factor is dispositive. Id. at 600. Determining whether an association is causal is a matter of scientific judgment, and scientists reliably applying the Bradford Hill factors may reasonably come to different conclusions about whether a causal inference may be drawn. Milward v. Acuity Specialty Products Group, Inc. , 639 F.3d 11, 18 (1st Cir. 2011) ; see also Ref. Man. at 553, 600.

An epidemiological study identifying a statistically significant association between the use of a drug and a particular adverse effect, accompanied by a reliable expert opinion that the association is causal, is "powerful" evidence of general causation. See Rider , 295 F.3d at 1198. The absence of epidemiological evidence, however, does not preclude admission of a general causation opinion in the Eleventh Circuit. See Kilpatrick v. Breg, Inc. , 613 F.3d 1329, 1336-37 (11th Cir. 2010) ; see also Rider , 295 F.3d at 1198-99 ; Wells v. Ortho Pharm. Corp. , 788 F.2d 741, 745 (11th Cir. 1986). Experts may rely on other, non-epidemiological evidence to prove causation; but where epidemiology is lacking, "the nature of the other evidence ... becomes that much more important, and [a] court's consideration of such evidence and the methodologies used must be that much more searching." See Kilpatrick , 613 F.3d at 1337 n.9.

b. Dose-Response Relationship

Another primary methodology for establishing causation is through evidence of a dose-response relationship, which is a "relationship in which a change in amount, intensity, or duration of exposure to [a drug] is associated with a change-either an increase or decrease-in risk of" adverse effects from that exposure. McClain v. Metabolife Int'l, Inc. , 401 F.3d 1233, 1242-43 (11th Cir. 2005). The relationship between dose and response is "the hallmark of basic toxicology" and the "single most important factor to consider" in evaluating the toxicity of a drug. Id. at 1242 ; see also Chapman , 766 F.3d at 1307. This is because virtually all substances have the potential to be harmful at high enough doses. See Chapman , 766 F.3d at 1307 ; see also Ref. Man. at 636. Inherent in this principle is the fact that, for the vast majority of substances, there are threshold doses below which no individual will respond and doses above which nearly everyone responds. See McClain , 401 F.3d at 1241-43. Consequently, a reliable expert opinion on general causation should address what levels of exposure to a drug increase the risk of adverse effects. See id. at 1241. Indeed, "[t]he expert who avoids or neglects this principle of toxic torts without justification casts suspicion on the reliability of his methodology." Kilpatrick , 613 F.3d at 1339 (quoting McClain , 401 F.3d at 1242 ).

c. Background Risk

A reliable methodology also should take into account the background risk for the disease at issue in the case. McClain , 401 F.3d at 1243. Background risk is the risk that members of the general public would have of developing the disease without exposure to the drug. Id. It encompasses all causes of the disease, whether known or unknown, except for the drug in question. Id. This is important because the aim of the other primary methodologies is to identify "agents that are associated with an increased risk of disease." See Ref. Man. at 552. An expert must know the background prevalence of a disease before he can determine whether the risk of that disease is increased as a result of exposure to the agent. See In re Denture Cream Products Liability Litigation , 795 F.Supp.2d 1345, 1355 (S.D. Fla. 2011), aff'd, Chapman , 766 F.3d 1296. Without background risk to establish a baseline, it is difficult to determine whether any incidence of a disease in individuals exposed to an agent is anything more than a coincidence. Chapman , 766 F.3d at 1308. Thus, a failure to identify or describe the background risk of a disease is a "serious methodological deficiency" and "substantial weakness" in an expert's general causation opinion. See id.

2. Secondary Methodologies

a. Biological Plausibility

Biological plausibility refers to a credible scientific explanation of the physiological processes or mechanisms by which a drug can cause a particular disease or adverse effect, based on current biological and pharmacological knowledge. See Ref. Man. at 604; see also McClain , 401 F.3d at 1253. Importantly, biological plausibility is not the same as biological certainty. See Daubert , 509 U.S. at 590, 113 S.Ct. 2786 ("Of course, it would be unreasonable to conclude that the subject of scientific testimony must be 'known' to a certainty; arguably, there are no certainties in science."); Jones v. Otis Elevator Co. , 861 F.2d 655, 662 (11th Cir. 1988) (stating that "absolute certainty is not required" from expert testimony). That is, an expert on biological plausibility need not definitively prove the biological means by which a drug acts in the body. See, e.g., In re Neurontin Mktg. Sales Practices & Prods. Liab. Litig. , 612 F.Supp.2d 116, 149 (D. Mass 2009) (finding that biological plausibility supported expert's opinion on causation despite the fact that there was "robust debate in the scientific community" on the proposed mechanism); In re PPA Prods. Liab. Litig. , 289 F.Supp.2d 1230, 1247 (W.D. Wash. 2003) ("The fact that the mechanism remains unclear does not call the reliability of the opinion into question."). Instead, a biological plausibility opinion is admissible so long as it is derived from and supported by reliable scientific knowledge and reasoning. See Allison , 184 F.3d at 1319, n.23 ("While scientific testimony need not be known to a certainty, Daubert does require that assertions be derived from scientific knowledge."); In re Seroquel Prods. Liab. Litig. , No. 6:06-md-1769, 2009 WL 3817866, *5 (M.D. Fla. Feb. 11, 2009) (finding biological plausibility opinion reliable where each step in expert's methodology had "ample scientific support" and was supported by "sound scientific reasoning"). Although biological plausibility, without more, cannot establish general causation; its existence "lends credence to an inference of causality" drawn from other, more substantial evidence. See Ref. Man. at 604; see also Chapman , 766 F.3d at 1308 ; Rider , 295 F.3d at 1202 ; Milward , 639 F.3d at 25-26.

b. Case Studies and Adverse Event Reports

Case studies document medical observations occurring coincident with the use of a prescription drug either by a single patient (a case report) or a small number of patients (a case series). Rider v. Sandoz Pharmaceuticals Corp. , 295 F.3d 1194, 1199 (2002). They tend to be brief recitals of clinical events and do not address prior medical history, use of other medications or drugs, risk factors, or the myriad of other issues necessary to scientifically evaluate whether the drug actually produced the observed adverse effect. See id. Moreover, case reports have no controls, are susceptible to inherent reporting biases, lack statistical context, and are not verifiable through meaningful peer review. See id. ; see also Kilpatrick , 613 F.3d at 1338. The difficulty with case reports is distinguishing between association and causation. For this reason, while case reports may supplement other evidence of causation, they cannot, standing alone, prove causation. See id.

One type of case report is more worthy of consideration in the general causation assessment, however. This report documents a patient's dechallenge and rechallenge events while taking a particular drug. A dechallenge event occurs where a patient's adverse side effects partially or completely disappear once the drug is stopped. Rider , 295 F.3d at 1199. If the side effects return when the patient resumes taking the drug, that is known as a rechallenge. See id. As other courts have noted, dechallenge and rechallenge data is "substantially more valuable than run-of-the-mill case reports because a patient's reactions are measured against his own prior reactions." Glastetter v. Novartis Pharm. Corp. , 252 F.3d 986, 990 (8th Cir. 2001) ; see also Rider , 295 F.3d at 1199 ("These reports, which may be analogized to controlled studies with one subject, can be particularly useful in determining whether a causal relationship exists."); Hollander v. Sandoz Pharmaceuticals Corp. , 289 F.3d 1193, 1212 (10th Cir. 2002) ; Giles v. Wyeth, Inc. , 500 F.Supp.2d 1048, 1051 n.7 (S.D. Ill. 2007). Nevertheless, dechallenge and rechallenge events "are still case reports and do not purport to offer definitive conclusions as to causation." Rider , 295 F.3d at 1200. Their "value is directly related to the degree of scientific control used in the" dechallenge or rechallenge exercise. See McClain , 401 F.3d at 1255. Thus, descriptions of dechallenge and rechallenge events generally are more reliable and probative of causation than a typical case report; however, alone, they cannot establish causation. See id.

Adverse event reports describe medical events that occurred during or after an individual's use of a prescription drug, which are submitted directly to the FDA by patients, healthcare professionals, and drug manufacturers. See 21 C.F.R. § 314.80(a). Generally speaking, in practice, the FDA adverse events reporting system (FAERS) simply entails "consumers call[ing] in to describe medical problems that they think they are experiencing from taking a product."

McClain , 401 F.3d at 1250. As a result, the system has several intrinsic limitations, including (1) uncertainty that the drug actually caused the reported event, since the FDA does not require that causation be proven before adverse event data is reported; (2) insufficient detail from which to evaluate causation; (3) information in the reports is unverified and subject to a variety of reporting biases; and (4) the underlying data may be affected by reporting bias stemming from publicity or litigation. See Ref. Man. at 731. Consequently, because they contain what amounts to "[u]ncontrolled anecdotal information," adverse event reports are generally considered "one of the least reliable sources" of support for a causation opinion. McClain , 401 F.3d at 1250.

c. In vivo and In vitro Studies

Toxicological knowledge often derives from in vivo studies, in which laboratory animals are exposed to a particular drug, with the outcomes monitored and compared to those for an unexposed control group. Ref. Man. at 639. In vivo studies offer a number of advantages, including that they can be conducted as true experiments, with exposure controlled and measured, they are replicable, they usually follow a generally accepted methodology, and they present fewer ethical limitations than human experimentation. See id. at 563; see also In re Paoli R.R. Yard PCB Litigation , 35 F.3d 717, 781 (3d Cir. 1994) ; In re Accutane Products Liability , 511 F.Supp.2d 1288, 1291 (M.D. Fla. 2007). However, the use of animal studies to prove causation in humans has "two significant disadvantages," which "are almost always fraught with considerable, and currently unresolvable, uncertainty." Ref. Man. at 563. First, extrapolating from animals to humans is difficult because biological "differences in absorption, metabolism, and other factors may result in interspecies variation in responses." Id. Second, most animal studies involve significantly higher doses of a drug than would ever be present in humans. Id. For these reasons, while animal studies may lend support to a general causation opinion, an expert must explain how and why the studies can be reliably extrapolated to prove comparable effects in humans. See Gen. Elec. Co. v. Joiner , 522 U.S. 136, 144-45, 118 S.Ct. 512, 139 L.Ed.2d 508 (1997) ; Kilpatrick , 613 F.3d at 1338-39 ; Rider , 295 F.3d at 1202.

These limitations apply with equal force to in vitro studies, which analyze the effects of drugs on human and animal cells, organs, or tissue cultures in a controlled laboratory setting. See Ref. Man. at 639. Observations about a drug's mechanism of action may be more readily gleaned from in vitro studies than from other sources, but the chemical reactions that occur in the artificial environment of a test tube or petri dish may differ from how the drug will react in, and impact, the complex biological system that is the human body. Ref. Man. at 564; Accutane , 511 F.Supp.2d at 1294-95. Thus, in vitro evidence alone cannot serve as a basis for a general causation opinion. See Kilpatrick , 613 F.3d at 1340-44. However, as with animal studies, in vitro data may be used to supplement other types of evidence, provided the expert explains how the in vitro data can be reliably extrapolated to predict a drug's effects in humans. See id.

d. Analogous Drugs

In analyzing causation, scientists sometimes draw from existing studies conducted on other drugs in the same class as, or which have a similar chemical structure to, the particular drug at issue in a case. See McClain , 401 F.3d at 1244-46 ; Rider , 295 F.3d at 1200-01. This approach is premised on the theory that drugs with similar chemical structures may be expected to have similar properties and produce analogous effects. See id. ; see also Richardson v. Richardson-Merrell, Inc. , 857 F.2d 823, 829 (D.C. Cir. 1988). Although such reasoning by analogy may have valid scientific uses, its value is somewhat limited in the context of establishing legal causation. This is because even within a given class of drugs, there may be "great chemical diversity" and those "minor deviations in chemical structure can radically change a particular substance's properties and propensities." See Rider , 295 F.3d at 1201 (quoting Glastetter , 252 F.3d at 990 ). Consequently, extrapolations from drugs within the same class may not support an expert opinion on general causation unless other reliable scientific evidence establishes the validity of the analogy. See McClain , 401 F.3d at 1246 ; Rider , 295 F.3d at 1200-01.

3. Weight of the Evidence

The preceding sections addressed the extent to which individual categories of scientific evidence may support an expert opinion on general causation in the Eleventh Circuit. In practice, however, many experts form a general causation opinion by weighing an entire body of scientific evidence. This "weight of the evidence" approach to analyzing causation can be considered reliable, provided the expert considers all available evidence carefully and explains how the relative weight of the various pieces of evidence led to his conclusion. See Milward , 639 F.3d at 17 ; see also In re Zoloft (Sertraline Hydrochloride) Prods. Liab. Litig. , 858 F.3d 787, 795-97 (3d Cir. 2017) (" Zoloft II ") aff'g 26 F.Supp.3d 449, 464 (E.D. Pa. 2014) ("Zoloft I "); Jones v. Novartis Pharm. Corp. , 235 F.Supp.3d 1244, 1272-73 (N.D. Ala. 2017) ; Magistrini v. One Hour Martinizing Dry Cleaning , 180 F.Supp.2d 584, 602 (D.N.J. 2002). Using this methodology, the expert must

(1) identify an association between an exposure and a disease, (2) consider a range of plausible explanations for the association, (3) rank the rival explanations according to their plausibility, (4) seek additional evidence to separate the more plausible from the less plausible explanations, (5) consider all of the relevant available evidence, and (6) integrate the evidence using professional judgment to come to a conclusion about the best explanation.

Milward , 639 F.3d at 17-18 ; Jones , 235 F.Supp.3d at 1273. Importantly, because the "weight of the evidence" approach involves substantial judgment on the part of the expert, it is crucial that the expert describe each step in the process by which he gathered and assessed the relevant scientific evidence. See Zoloft II , 858 F.3d at 795-97 ; In re Seroquel Products Liability Litigation , No. 6:06-md-1769, 2009 WL 3806435, at *4-6 (M.D. Fla. June 23, 2009). To be considered reliable, the expert's weighing process must have been "based on methods and procedures of science, rather than on subjective belief or unsupported speculation." Zoloft II , 858 F.3d at 796. Otherwise, the methodology amounts to nothing more than the expert's ipse dixit , which the Supreme Court has admonished district courts against admitting into evidence. See Joiner , 522 U.S. at 146, 118 S.Ct. 512. Moreover, an expert cannot merely aggregate various categories of otherwise unreliable evidence to form a reliable theory of general causation. See Siharath v. Sandoz Pharm. Corp. , 131 F.Supp.2d 1347, 1371 (N.D. Ga. 2001) (cautioning that expert "cannot lump together lots of hollow evidence" to establish medical causation); see also Hollander , 289 F.3d at 1216 ("To suggest that those individual categories of evidence deemed unreliable by the district court may be added to form a reliable theory would be to abandon the level of intellectual rigor of the expert in the field."); Glastetter , 252 F.3d at 992 (deciding that neither individual items nor an aggregate of evidence provided a reliable scientific basis for experts' conclusions). Instead, every aspect of the expert's analysis-including his methodology, the combination of facts and scientific evidence on which he relies, and the links between the evidence and his conclusions-must be shown to satisfy Rule 702 and Daubert. See McClain , 401 F.3d at 1245 ; see also Heller v. Shaw Industries, Inc. , 167 F.3d 146, 155 (3d Cir. 1999). Where an expert is found to have applied the "weight of the evidence" approach with "the same level of intellectual rigor" used by experts in the field, see Kumho Tire, at 152, 119 S.Ct. 1167, his general causation opinion typically will be deemed reliable and admissible. See Milward , 639 F.3d at 17 ; see also Zoloft II , 858 F.3d at 795-97 ; Jones , 235 F.Supp.3d at 1272-73 ; Magistrini , 180 F.Supp.2d at 602.

C. Reliability of Common Evidence of General Causation

Plaintiffs' experts, each to a greater or lesser extent, rely on much of the same evidence to conclude that Abilify can cause impulsive gambling and other impulse control disorders. The Court addresses the reliability of the common evidence together in this section. In Section II(D), the Court addresses Defendants' expert-specific objections to Drs. Bechara, Glenmullen, Hollander, Luepker, and Madigan.

1. Epidemiological Evidence-The Etminan Study

Three of Plaintiffs' experts-Drs. Glenmullen, Hollander, and Madigan-base their opinions, in part, on an epidemiological study published by Dr. Mahyar Etminan and Dr. Ric M. Procyshyn in February 2017, in which a statistically significant association was found to exist between Abilify and impulse control disorder, and also between Abilify and gambling disorder ("Etminan Study"). See Mahyar Etminan, Risk of Gambling Disorder and Impulse Control Disorder with Aripiprazole, Pramipexole, and Ropinirole , 37 J. CLINICAL PSYCHOPHARMACOLOGY 1 (2017), ECF No. 428-13. The Etminan Study is the only epidemiological study conducted to date that analyzes whether Abilify is associated with an increased risk of gambling and impulse control disorder. Defendants challenge the reliability of the Etminan Study on multiple grounds, arguing that it is "so riddled with flaws as to be inherently unreliable." See ECF No. DSJ, 428-26 at 45. The Court disagrees and, for the reasons that follow, finds the Etminan Study sufficiently reliable to support an expert opinion on general causation in this case.

The Etminan Study is an epidemiological case-control study in which the authors analyzed medical and pharmaceutical billing information for over six million individuals, drawn from a large insurance claims database known as LifeLink. See Etminan Study, ECF No. 428-13 at 1. The database included, inter alia , patients' diagnoses, as identified by ICD-9-CM Codes, and all prescriptions they filled between 2006 and 2014. See id. Within this data, the authors first identified all individuals whose insurance records reflected a diagnostic code for either pathological gambling or impulse control disorder. These individuals served as the Etminan Study's "case" group. Next, from the same data, the authors drew a random sample of similar individuals whose records contained neither diagnostic code. These individuals served as "controls." The authors then compared the cases (individuals diagnosed with pathological gambling or impulse control disorders ) to the controls (individuals with no such diagnoses) based on the prevalence of exposure to Abilify in each group. Exposure to Abilify was defined for the cases as one prescription for Abilify having been filled during the year before the pathological gambling or impulse control disorder diagnosis, and in corresponding calendar time for the controls. The Study found that individuals exposed to Abilify had a statistically significant higher incidence of pathological gambling and impulse control disorder diagnoses than did unexposed individuals.

The Etminan Study described the existence and strength of the association found between Abilify, pathological gambling, and impulse control disorder in the random sample from the LifeLink database in terms of "rate ratios," also known as relative risk. Relative risk is simply a comparison of the incidence of a disease in exposed individuals with its incidence in unexposed individuals. See Ref. Man. at 566. A relative risk of 1.0 means there is no difference in risk between the exposed and unexposed groups; in other words, there is no association between exposure to the drug and the disease. See Ref. Man. at 567; see also Allison , 184 F.3d at 1315 n.16. A relative risk above 1.0 indicates an increased risk in the exposed group, see Ref. Man. at 567, and "[r]isks greater than 2.0 permit an inference that the [disease] was more likely than not caused by the [drug]," see Allison , 184 F.3d at 1315 n.16. Relative risk estimates are often accompanied by a "confidence interval," which provides, in essence, a margin of error. See Ref. Man. 579-80. Confidence intervals identify the range of likely values, on either side of the relative risk estimate for a population sample, that would be expected to encompass the results a specified percentage of the time (e.g. , 95%) if random samples were repeatedly drawn from the same population as the subject study. See id. at 580. Importantly, if the confidence interval contains the value 1.0 or less, then the results of the study are not considered statistically significant. On the other hand, if the lower bound of the confidence interval exceeds 1.0, then the results are considered statistically significant.

In this case, the Etminan Study reported a relative risk of 5.23 for pathological gambling in individuals exposed to Abilify as compared to unexposed individuals, with a 95% confidence interval of 1.78-15.38. See Etminan Study, ECF No. 428-13 at 3. This means that the Study predicted that the increased risk of pathological gambling for Abilify patients within any given sample of the entire LifeLink database would likely fall anywhere between 1.78-15.38. Because the lower bound of the confidence interval (1.78) exceeds 1.0, this is statistically significant. The Study also reported a relative risk of 7.71 for impulse control disorder, with a 95% confidence interval of 5.81 and 10.34. This too is statistically significant. Finally, an analysis restricted to patients with bipolar disorder alone yielded a relative risk of 3.38 for pathological gambling in Abilify patients, with a 95% confidence interval of 1.68-8.48, which is also statistically significant. Defendants do not dispute the accuracy of the Etminan Study's relative risk and confidence interval calculations.

Plaintiffs' biostatistician, Dr. David Madigan, analyzed the Etminan Study and found it to be "methodologically sound" with a "highly statistically significant result." , See Madigan Rep., ECF No. 427-1 at 30. This conclusion was based, in part, on the "strong" and "very substantial" relative risk figures reported in the Study; again, numbers that Defendants do not dispute. See Madigan Tr., ECF No. 596-4 at 49. Dr. Madigan also calculated a p -value for the Study's relative risk finding for pathological gambling. A p -value is a separate, widely established indicator of statistical significance, which measures the probability of obtaining the observed results-in this case, the increased risk of developing pathological gambling with exposure to Abilify-if, in reality, there is no true association between the drug and the adverse effect. See Ref. Man. at 249-50, 576-77; see also Matrixx Initiatives, Inc. v. Siracusano , 563 U.S. 27, 39 n.6, 131 S.Ct. 1309, 179 L.Ed.2d 398 (2011). Stated differently, the p -value provides an estimate of the probability that chance alone produced the observed association between the drug and the adverse effect. See id. The lower the p -value, the less likely it is that the observed result can be explained by chance alone. See Ref. Man. at 250-51; see also Matrixx , 563 U.S. at 39 n.6, 131 S.Ct. 1309. Generally, p -values are considered statistically significant where they are less than or equal to .05 (p = 5%). See Ref. Man. at 251; see also Eastland v. Tenn. Valley Auth. , 704 F.2d 613, 622 (11th Cir. 1983) ("Generally ... a probability level of .05 is accepted as statistically significant."). By Dr. Madigan's calculation, the p -value for pathological gambling in the Etminan Study is .002, which is well below the traditional threshold for statistical significance. See Madigan Rep., ECF No. 427-1 at 25. According to Dr. Madigan, this p -value indicates that the probability of the Etminan Study producing a 5.23-fold increased risk of pathological gambling by chance alone is one in 500. See Madigan Tr., ECF No. 596-4 at 49. Defendants do not dispute the accuracy of this calculation.

Dr. Madigan discussed, at length, the strengths and limitations of case-control studies generally, as well as those of the Etminan Study specifically. See Madigan Rep., ECF No. 427-1 at 21-25, Madigan Supp., ECF No. 427-1 at 85-90; Madigan Tr., ECF No. 596-4 at 42-47, 54-58. In particular, with respect to the potential effect of bias on the Study's results, Dr. Madigan explained that the relative risk calculations are simply too "substantial" and "robust" to be explained by investigator bias. See Madigan Tr., ECF No. 596-4 at 70. In short, Dr. Madigan opines that a case-control study is "highly unlikely" to yield increased risk estimates like those found in the Etminan Study "in the absence of a true" association. See Madigan Supp., ECF No. 427-1 at 91. In his view, the FDA called for a case-control study "clarify[ing]" the association between Abilify and impulse control disorders, and the Etminan Study reliably did exactly that. See Madigan Tr., ECF No. 596-4 at 71.

Defendants argue that numerous methodological flaws render the Etminan Study unreliable under Rule 702 and Daubert , including a deficient study design, failure to account for the risk of confounders, and the presence of bias. They also challenge Dr. Madigan's defense of the Etminan Study, which they claim is untenable in light of his prior published research criticizing both healthcare database research and the use of p -values as a measure of statistical significance. The Court addresses each category of objections in turn.

a. Study Design

Defendants criticize the Etminan Study's use of the LifeLink database because the database was not designed for research purposes. This criticism has little, if any, merit. The use of health insurance claims databases for epidemiologic research is well-supported by the medical literature, which is an important consideration under Daubert. See Daubert , 509 U.S. at 589-90, 113 S.Ct. 2786. Their "representativeness [of routine clinical practice], large size, and capacity to contain large quantities of [long-term] clinical data on each patient" can make them a "powerful tool" for studying the use, efficacy, and safety of prescription drugs. Schneeweiss, 48 J. CLIN. EPIDEMIOLOGY at 334, DX-122 at 12. Indeed, large databases are particularly advantageous for studying relatively rare adverse effects of a drug, as in this case, or where multiple possible adverse effects are of interest. See id. at 325, DX-122 at 3.

With that said, large database research is not without limitations, one of which is the unavailability of medical records to confirm the accuracy of the data and to provide potentially significant clinical information not reported in the database. Defendants argue that this limitation is fatal to the Etminan Study's reliability under Daubert. The Court disagrees. While it is true that the medical literature encourages record review, the medical community also recognizes that health information privacy laws have "constrained the availability of" individual medical records "for uses other than the direct care of patients." See id. at 327-28, DX-122 at 5-6. Large database research for pharmacovigilance purposes is generally accepted in the scientific community, see supra n.34, and has been found to be a reliable methodology by other courts, even where no medical record review occurred. See Rheinfrank v. Abbott Labs., Inc. , No. 1:13-cv-133, 2015 WL 13022172, at *13 (S.D. Ohio Oct. 2, 2015) (finding "data-mining" of the FDA adverse event reporting system database a reliable methodology for determining whether a signal for developmental delay from in utero exposure to Depakote existed); In re Fosamax (Alendronate Sodium) Prods. Liab. Litig. , No. 3:11-cv-05304, 2013 WL 1558690, at *8 (D.N.J. April 10, 2013) ("[D]ata mining in pharmacovigilance[ ] is generally accepted in the scientific community and has become routine both in the pharmaceutical industry and amongst regulators worldwide.").

The LifeLink database, despite Defendants' criticisms, contains a sufficiently comprehensive dataset of patients, medical diagnoses and prescription claims to reliably serve the epidemiological objectives of the Etminan Study. Indeed, "claims data of this type provide some of the best data on drug exposure in pharmacoepidemiology." Brian L. Strom, Overview of Automated Databased in Pharmacoepidemiology , in PHARMACOEPIDEMIOLOGY 158, 159 (Brian L. Strom et al. eds., 5th ed. 2012), DX-129 at 2. The Etminan Study's statistical analysis of the LifeLink data is capable of being tested, and the Study itself has been subjected to peer review and publication in a reputable medical journal. This is all that Rule 702 and Daubert require. See Chapman , 766 F.3d at 1305 (citing Daubert , 509 U.S. at 593-94, 113 S.Ct. 2786 ). The alleged inadequacies of the LifeLink database may impact the weight afforded to the Etminan Study's conclusions, but not its reliability or admissibility under Daubert.

Defendants' next argument, which relates to the database challenges addressed above, is that the Etminan Study is unreliable for its inability to confirm that individual patients in the LifeLink database were ever actually exposed to Abilify ; that is, that they actually took the Abilify they were prescribed. The Study did not attempt to validate medication usage, even though its lead author, Dr. Etminan, has done so in other epidemiological studies. This criticism also fails.

All epidemiological studies that make use of large healthcare databases are vulnerable to the risk of drug exposure misclassification, which is the risk of inaccurately measuring actual exposure to a drug. See Schneeweiss, 48 J. CLIN. EPIDEMIOLOGY at 328, DX-122 at 6. This is because claims databases only reflect the dispensing of medications and not actual medication use. See id. Despite this limitation, the use of pharmacy dispensing data as a proxy for drug usage is seen in the scientific community as "the gold standard of drug exposure information compared with self-reported information or prescribing records in outpatient medical records." See id.; see also Strom at 159 ("[C]laims data of this type provide[s] some of the best data on drug exposure in pharmacoepidemiology."), DX-129 at 2. Short of physically monitoring ingestion or requiring study subjects to undergo routine laboratory testing to ascertain medication levels, there appears to be no more reliable means of measuring drug exposure than pharmacy claims data.

The fact that the Etminan Study did not attempt to correct for the risk of drug exposure misclassification does not render it unreliable under Daubert. There is no evidence in the record of an established epidemiological protocol for addressing drug exposure misclassification concerns. See Kumho Tire, at 152, 119 S.Ct. 1167 (an expert in the courtroom must use "the same level of intellectual rigor" that characterizes the practice of an expert in the relevant field). There also is no evidence that Dr. Etminan violated his own methodological standards with respect to this issue. While Defendants are correct that Dr. Etminan has, in the past, tried to "control" for the risk of drug exposure misclassification in different ways, there are also numerous published, peer-reviewed studies in which Dr. Etminan's treatment of the issue mirrors that used in the Etminan Study. There is no evidence that these studies have been criticized in the scientific community for failing to account for the risk of exposure misclassification in the study design. Moreover, the record in this case suggests that at least one of Dr. Etminan's prior validation techniques-using two prescriptions as a proxy for exposure, instead of a single prescription, see supra n.36-might have negatively skewed the Etminan Study results, if used. Both Defendants' expert, Dr. Marc Potenza, and Plaintiffs' expert, Dr. Eric Hollander, cautioned that a study utilizing two Abilify prescriptions as a proxy for exposure could miss individuals who developed symptomology during their first prescription and who subsequently discontinued the drug. See Potenza Tr., ECF No. 596-7 at 84-85; Hollander Tr., ECF No. 596-4 at 116. Under these circumstances, the Etminan Study cannot be considered unreliable for failing to control for the risk of drug exposure misclassification. This objection may be probative on the weight of the Study, but not its admissibility. See Bazemore v. Friday , 478 U.S. 385, 400, 106 S.Ct. 3000, 92 L.Ed.2d 315 (1986) ("Normally, failure to include variables will affect the analysis' probativeness, not its admissibility.").

Defendants' next argument with respect to the Etminan Study's design is that the Study cannot reliably measure the incidence of iatrogenic gambling (i.e. , medication-induced) in Abilify patients because it identified cases of gambling disorder in the LifeLink database using medical billing codes that are based on the DSM-5 diagnostic criteria for idiopathic gambling (i.e. , gambling disorder that occurs spontaneously and with no known cause). The Court disagrees. The LifeLink database classifies diagnoses according to ICD-9-CM codes, not the DSM-5 diagnostic criteria. See Etminan Study, ECF No. 428-13 at 1. Although the drafters of these two classification systems have, in recent years, attempted to "harmonize [them] as much as possible," the ICD-9-CM and DSM-5 are not identical. See DSM-5 at 11. Indeed, the two publications serve different purposes and, in some circumstances, diverge or are discordant with one another. See id. The Court finds that this case presents one of those circumstances. The DSM-5 expressly recognizes the existence of iatrogenic gambling, but excludes the condition from its diagnostic criteria for "gambling disorder." See DSM-5 at 589, ECF No. 428-3 at 68. In other words, the DSM-5 acknowledges the condition of iatrogenic gambling and explains that the diagnostic criteria for idiopathic gambling do not apply to it. See id. In contrast, the ICD-9-CM definition of "pathological gambling" does not appear to be limited, either explicitly or implicitly, to idiopathic gambling. See ICD-9-CM, PX-063 at 1. By its terms, the ICD-9-CM diagnostic code encompasses any "preoccupation with gambling and the excitement that gambling with increased risk provides," even where "it may lead [the patient] to lie, steal, or lose a significant relationship, job, or educational opportunity." See id. There simply is no carve out for, or exclusion of, iatrogenic gambling (medication-induced) from the ICD-9-CM definition. See id. Thus, it would be reasonable for a scientist to conclude that "pathological gambling" under the ICD-9-CM includes iatrogenic gambling. In turn, the Etminan Study's use of that same ICD-9-CM code to identify cases of pathological gambling in the LifeLink database is likewise reasonable. This challenge does not undermine the reliability or admissibility of the Etminan Study.

Defendants' last argument is that the Etminan Study is unreliable because the time between exposure to Abilify and the diagnoses of pathological gambling in the random sample taken from the LifeLink database was too short to be compatible with a cause-effect relationship. This argument is based on the Study's finding that five patients were exposed to Abilify in the year preceding their diagnoses of pathological gambling, with an average, or mean, time to diagnosis of 20 days and a standard deviation of 17.4 days. See Etminan Study, ECF No. 428-13 at 3. The standard deviation is a measure of statistical dispersion; that is, the average distance between the five individual time-to-diagnosis data points and the mean. See Ref. Man. at 239. None of the five actual time-to-diagnosis periods was individually reported in the Study. See id. However, Defendants claim that the standard deviation in the Study indicates that at least one of the five patients was diagnosed with pathological gambling within three days of his exposure to Abilify, which even Dr. Etminan, at his deposition, agreed was "unlikely." See Etminan Tr., ECF No. 427-3 at 80; see also Def. Madigan Motion, ECF No. 427-20 at 9. Defendants insist that, under the DSM-5 criteria, gambling disorder "takes up to twelve months to develop into a disease," so if the five diagnoses in the Etminan Study followed this criterion, the patients' pathological gambling necessarily would have preceded their exposure to Abilify. DSJ, ECF No. 428-26 at 27.

This criticism is not fatal to the Study's reliability under Daubert for several reasons. First, the standard deviation of 17.4 days does not dictate a conclusion that there must have been a three-day period between exposure and diagnosis for at least one of the five patients in the random sample analyzed in the Study. At his deposition, Dr. Etminan, the Study's lead author, testified that while the time-to-diagnosis for one of the patients could have been "a matter of days," he could "not [be] sure exactly what" the actual times-to-diagnosis were for any of the five patients "without having the data" from the LifeLink database to review. See Etminan Dep., ECF No. 457-7 at 42. Dr. Madigan, who is the only statistics expert in this case, calculated at least two possible distributions of the five individual time-to-diagnosis periods, given the mean time of 20 days and the standard deviation of 17.4 days: (1) a distribution of 3, 41, 13, 37, and 8 days, respectively; or (2) a distribution of 50, 16, 8, 8, and 17 days. See Madigan PPT, PX-051 at 22; Madigan Rep., ECF No. 427-1 at 24. According to Dr. Madigan, there is no way to determine, from the information reported in the Study, which of these two possible distributions represents the actual distribution of time-to-diagnosis periods for the five patients found to have been exposed to Abilify within the year preceding their diagnosis of pathological gambling in the LifeLink database. See Madigan Tr., ECF No. 596-4 at 62-63. None of Defendants' experts disputed Dr. Madigan's calculations. Thus, the evidence shows that the minimum time-to-diagnosis in the Etminan Study could have been as few as three, or as many as eight, days.

Second, the possibility of pathological gambling or other impulse control symptoms developing within either three or eight days of exposure to Abilify is consistent with the scientific literature. Multiple published case reports