First, do no harm. Although this dictum is not in the Hippocratic Oath, it has become a central tenet that guides the practice of medicine. However, our profession has fallen well short of achieving this lofty ideal. Iatrogenic contributions to our patients' woes have long been recognized as an important issue. Recently, the Institute of Medicine's committee on patient safety issues published extensive recommendations for preventing medical errors (
1). Drug-drug interactions (DDIs) represent a critically important and widely underrecognized source of medical errors.
DDIs arise in numerous ways. A wide array of pharmacodynamic interactions exists, in which the receptor effects of different agents interact to produce synergy or cancellation of drug effects. A similarly wide range of pharmacokinetic interactions exist, in which the blood levels of given agents may be raised or lowered. We have long grappled with the inherent complexity posed by our vast armamentarium of drugs and their numerous and varied effects. However, only in the past few years have we gained an appreciation for the prevalence, complexity, and clinical importance of two key systems that significantly influence drug levels, namely the cytochrome P-450 system and the P-glycoprotein transporter. Many clinicians tend to dismiss these concerns as overly academic or irrelevant to actual patient care. In fact, DDIs have an enormous impact on patient care, and the pervasively poor recognition of DDIs is a major part of the problem.
DDIs are an especially important issue for psychiatrists, because many commonly used psychotropic agents have abundant P-450 and P-glycoprotein effects that may affect the blood levels of other drugs (
2,
3). Similarly, many of the nonpsychotropic drugs that our patients take can significantly increase or decrease the blood levels of most psychotropic agents. To treat patients in a competent and safe manner, some awareness of the DDI issue and some means of detecting DDIs are essential.
Frequency and recognition
In one landmark, often-cited study by Grymonpre and colleagues (
4) that was published in 1988, DDIs were found to be the main cause for roughly 2.8 percent of all admissions among persons older than 50 years who were taking medications. In retrospect, with our current greater ability to recognize DDIs, it is likely that this figure is higher. In another study, a retrospective review of the records of 205 patients consecutively admitted into an emergency department who either received three or more medications or were older than 50 years and received two or more medications revealed significant potential DDIs among 47 percent of these patients. Half the patients with potential DDIs went to the emergency department for treatment because they were experiencing an adverse event arising from a DDI (
5). The study also found that the risk of a potential DDI rose from 13 percent for patients who took two drugs to 82 percent for patients who took seven or more.
Other studies have examined and quantified the risk of DDIs for specific treatments. For example, in one study of 104 patients who were treated with warfarin, significant increases were found in length of hospital stay (a mean increase of 3.14 days) and in prothrombin-time results (a mean increase of 24 percent) among those who received an interacting drug (55 percent), compared with those who did not (
6).
In a large Canadian study of 28,705 patients who were taking HMG-CoA reductase inhibitors (statins), the group of patients who were taking drugs that interacted with the particular statin that they were taking (25.9 percent of the total cohort) experienced significantly greater rates of hospital admissions, physician visits, and total health care costs than those who were not taking such drugs (
7).
A more recent study of patients aged 66 years or older who were treated with glyburide, digoxin, or an angiotensin-converting enzyme inhibitor and were hospitalized because of hypoglycemia, digoxin toxicity, or hyperkalemia, respectively, revealed that in the week before admission the patients were six times as likely to have been exposed to co-trimoxazole, 12 times as likely to have been exposed to clarithromycin, and 20 times as likely to have been exposed to a potassium-sparing diuretic. These specific DDIs accounted for 3.3 percent of all hypoglycemia admissions, 2.3 percent of all digoxin-toxicity admissions, and 7.8 percent of all hyperkalemia admissions (
8). These figures are even more daunting when one considers that these specific interacting drugs represent only a small subset of drugs that could easily interact with the index drugs to produce the above complications.
Recognition and detection of DDIs by physicians is generally poor. In one study of 263 physicians who practiced in the Southern California Department of Veterans Affairs system (21 percent of the physicians surveyed were psychiatrists), clinicians recognized only 53 percent of DDIs of moderate to severe intensity and only 54 percent of potentially fatal combinations, such as sertraline with phenelzine or sildenafil with isosorbide (
9). In a study by Langdorf and colleagues (
10), eight physicians who worked in the emergency department missed 21 percent of those DDIs that necessitated a change in clinical management.
I've spoken at length with clinicians of many different disciplines—surgeons, internists, fellow psychiatrists, and others—who have adopted the attitude that it is the responsibility of the pharmacist who dispenses the medications to detect a DDI before the patient receives the medication. This stance does not meet the standard of care and would be grounds for malpractice. Apart from this practical consideration, it is also unreliable to depend only on the pharmacist. Even armed with the pharmacy computer, pharmacists make mistakes with alarming frequency.
In one particularly striking series Cavuto and colleagues (
11) visited 50 pharmacies in the Washington, D.C., area and attempted to fill prescriptions simultaneously for erythromycin and terfenadine. This study occurred at a time when the risk of developing a torsades de pointes arrhythmia with this combination of drugs was well recognized but terfenadine had not yet been withdrawn from the U.S. market. Sixteen of the 50 pharmacies (32 percent) readily filled the prescriptions. Even when these 16 pharmacists were explicitly asked, "Is there any problem with taking these two medications together?" nine said that they could be taken together. Forty-eight of the 50 pharmacies were using computer programs designed to prevent DDIs.
Clinical importance
Several of the aforementioned studies demonstrate how DDIs lead to increased rates of hospitalization, increased lengths of stay, and patient morbidity (
4,
5,
6,
7,
8). Other examples include the study by Hamilton and colleagues (
12), in which the odds ratios for hospitalization among 157,542 Medicaid patients who were exposed to known DDIs were compared with the odds ratios for 157,542 Medicaid patients who were exposed to only one of the agents in the DDI pair. The study found 23 DDIs that were significantly associated with hospitalization. Several of the most potently predisposing combinations included rifampin with azole antifungals, barbiturates with glucocorticoids, selective serotonin reuptake inhibitors (SSRIs) with phenytoin, SSRIs with warfarin, and sulfonamides with warfarin. Also, a large study of 13,062 Kaiser Permanente enrollees found that patients who received therapeutically risky drug combinations were 34 percent more likely to be hospitalized (
13).
Although no comprehensive prospective or even retrospective study has examined patient mortality as a result of DDIs, this concern led the Food and Drug Administration to remove terfenadine, astemizole, and cisapride from the U.S. market in the past seven years (
14). When combined with one of the many potent inhibitors of P-450 3A4, each of these agents is capable of producing torsades de pointes, which can progress to ventricular fibrillation and death (
15,
16). Although a detailed examination of the mechanisms of DDIs is beyond the scope of this column, nothing is unique about the mechanisms that led to these deaths. Any agent with a low to medium therapeutic index (LD
50 [lethal dose 50%]/ED
50 [effective does 50%]) can have its blood level dangerously increased through a DDI.
Cost
Accepting the statistic from Grymonpre and colleagues (
4) that DDIs cause roughly 2.8 percent of all hospitalizations, Hamilton and colleagues (
12) stated, "Using a cost-of-illness model, this could represent 245,280 hospital admissions/year, costing the health care system $1.3 billion." In one isolated example, Shad and colleagues (
17) reported on the case of a 67-year-old woman with a history of Parkinson's disease who had been receiving selegiline and trihexyphenidyl. Fluoxetine was added to treat depressive symptoms, resulting in delirium and a protracted hospital course that lasted 15 days and incurred a total cost of $17,213. These studies demonstrate that DDIs lead to increased rates of hospitalization and increased lengths of stay and obviously imply that increased costs are associated with these complications. What is harder to quantify is the opportunity cost of lost work hours of patients and those who must care for them, as well as the decreased productivity that arises from the stress that these DDIs create among patients and caretakers. As with all other domains of clinical medicine, the biopsychosocial model applies to the sequelae of DDIs. However, the lives lost to DDIs give us more urgent reasons to find solutions than the tangible and intangible health care costs.
Prevention
Preventing DDIs relies on two linked functions: detection and recognition. The most important developments in our ability to detect DDIs involve computer programs. Several medical systems have already demonstrated that the use of computers can lead to significant decreases in overall medical errors, including DDIs. The installation of the Partners HealthCare System's provider order entry system, employed at Massachusetts General Hospital and its affiliates, was associated with a 55 percent decrease in serious medical errors (
18). In one large Israeli study of 775,186 clinic patients who were monitored for 18 months, the rate of patient exposure to severe potential DDIs declined 62.8 percent after the implementation of a computerized drug alert system (
19).
However, computer programs are not a panacea. Hazlet and colleagues (
20) evaluated the ability of nine different DDI software programs to detect 16 well-established DDIs that were contained within six different fictitious patients' medication regimens. The sensitivities of these programs ranged from .44 to .88 (with 1 representing perfect), and their specificities ranged from .71 to 1, not exactly a gold standard. Also, in the study by Langdorf and colleagues (
10) the percentage of physician-identified, computer-missed DDIs that were clinically significant was 45 percent. Again, this result implies that there is a sensitivity problem with some DDI software. This finding should not be altogether surprising, because the sensitivity of a DDI program is limited by the completeness of the database that is entered into it, and entering all known DDIs is a dauntingly voluminous task.
Future directions
Although the detection of DDIs with computer programs has not been perfected, the more difficult problem is that of recognition of DDIs by physicians, even when they are equipped with such programs. The two largest challenges in this respect are the limitations of human memory and alert fatigue, which means habituation to computer alerts to the point that significant content may often be overlooked. Thus it has developed that "passive" DDI programs are often very complete and detailed, but their usefulness is limited in that the clinician must first realize the potential for a DDI and the need to consult the program. This system is by no means reliable. However, more "active" programs, in which patients' regimens are already uploaded and computerized order entry triggers DDI alerts, are often stripped of most of their completeness and complexity to avoid what is perceived as unacceptable obstruction to physician workflow.
So what is to be done? I am hopeful that electronic medical records and order entry are the wave of the future and that "active" computer programs will grow more sophisticated, providing graded magnitudes of alert status with multileveled options for learning more about a potential DDI. In the meantime I offer these pieces of advice. First, become familiar with the DDIs for the drugs you use most frequently. Second, pay special attention to potential DDIs when prescribing agents with a low therapeutic index. Third, refer frequently to tables, charts, references, or computer programs that you like and trust and keep them handy. Fourth, encourage your patients to get all their medications at the same pharmacy and to enroll in that pharmacy's DDI monitoring program. Fifth, whenever possible, try to select agents with a low likelihood of producing DDIs within a given class of agents, such as citalopram and escitalopram among the SSRIs, pravastatin among the statins, and azithromycin among the macrolides. Hopefully, these prudent measures will provide reasonable protection from the worst that DDIs have to offer until computer software is able to negotiate the delicate balance between completeness and utility.