Drug Interaction Severity Levels Explained: From Minor to Contraindicated

Not all drug interactions are created equal. A minor pharmacokinetic interaction that slightly raises a drug's plasma concentration is fundamentally different from a contraindicated combination that can cause fatal cardiac arrhythmias. Yet many drug interaction databases and APIs return results as flat lists, leaving the consuming application to figure out which interactions actually require clinical action.

This gap between detection and actionability is the root cause of alert fatigue, the single most cited reason clinicians override drug interaction warnings. Studies consistently show that clinicians override 90 to 96 percent of drug interaction alerts in electronic health record systems. The problem is not that the alerts are wrong. It is that they are presented without meaningful prioritization, making it impossible to distinguish a life-threatening contraindication from a minor monitoring suggestion in the moment of prescribing.

Severity classification is the mechanism that transforms a raw interaction detection into a clinically useful signal. When an API returns not just the fact that two drugs interact but how seriously they interact, the consuming application can make intelligent decisions about alert presentation, workflow interruption, and documentation requirements.

RxLabelGuard assigns one of five severity levels to each detected drug interaction. These levels form a clear hierarchy from lowest to highest clinical significance, and each level maps to specific implications for how the interaction should be handled in a clinical workflow.

The five levels, from most to least severe, are: contraindicated, major, moderate, minor, and unknown. This classification system aligns with the severity tiers used by the major clinical pharmacology references and commercial drug databases, ensuring that RxLabelGuard output is immediately interpretable by systems already built around these categories.

A contraindicated interaction represents a drug combination that should never be used together under any circumstances. The risk of serious harm or death is well established, and no clinical benefit justifies the combination. FDA labeling typically uses explicit language such as 'contraindicated,' 'must not be used concomitantly,' or 'co-administration is contraindicated' when describing these interactions.

Classic examples include the combination of methotrexate with live vaccines, where the immunosuppressive effect of methotrexate can lead to disseminated vaccine infection. Another textbook case is the concurrent use of MAO inhibitors with serotonergic drugs like meperidine, which can precipitate serotonin syndrome with potentially fatal hyperthermia and neuromuscular rigidity.

In a clinical decision support system, contraindicated interactions should generate hard-stop alerts that require explicit clinical override with documented justification. These are the interactions where the alert must interrupt the workflow, regardless of the clinician's tolerance for interruptions.

Major interactions carry a high risk of serious adverse outcomes, including hospitalization, permanent injury, or death, but they differ from contraindicated combinations in that the drugs may sometimes be used together when the benefit clearly outweighs the risk and appropriate monitoring is in place.

The combination of warfarin and nonsteroidal anti-inflammatory drugs is a frequently encountered major interaction. NSAIDs increase bleeding risk through multiple mechanisms: they inhibit platelet function, can cause gastric erosion, and some NSAIDs displace warfarin from protein binding sites, raising free warfarin levels. The interaction is well documented and clinically significant, but a rheumatologist managing a patient with both atrial fibrillation and severe inflammatory arthritis may determine that the combination is necessary with close INR monitoring.

Another example is the combination of QT-prolonging drugs, such as concurrent use of fluoroquinolone antibiotics with antiarrhythmic agents. The additive QT prolongation can trigger torsades de pointes, a potentially fatal ventricular arrhythmia. ECG monitoring and electrolyte correction can mitigate but not eliminate this risk.

For CDS implementations, major interactions typically warrant interruptive alerts with a lower override threshold than contraindicated pairs. The alert should clearly communicate the specific risk and recommended monitoring, enabling an informed clinical decision rather than demanding absolute avoidance.

Moderate interactions have the potential to worsen a patient's condition or require therapeutic adjustment, but the risk of serious harm is lower than with major interactions. These interactions often manifest as changes in drug efficacy or predictable side effect amplification that can be managed with dose modification, timing changes, or routine monitoring.

A common example is the interaction between ACE inhibitors and potassium-sparing diuretics like spironolactone. Both drugs increase serum potassium levels, and the combination raises the risk of hyperkalemia. This interaction is clinically significant and requires potassium monitoring, but it is also an intentional therapeutic combination in heart failure management where the benefits are well established.

Statins combined with certain calcium channel blockers present another moderate interaction. Diltiazem and verapamil inhibit CYP3A4 metabolism of statins like simvastatin and atorvastatin, increasing statin exposure and the risk of myopathy. The standard management is to use a lower statin dose or switch to a statin not metabolized by CYP3A4, such as rosuvastatin.

Moderate interactions are where alert fatigue becomes most problematic. These combinations are common in clinical practice, often intentional, and usually manageable. CDS systems that present moderate interactions with the same urgency as contraindicated pairs train clinicians to override all alerts indiscriminately. The solution is tiered presentation: moderate interactions should be visible but not workflow-blocking, perhaps appearing as informational banners or documentation prompts rather than hard stops.

Minor interactions have limited clinical significance and rarely require changes to therapy. They may produce small pharmacokinetic changes that are unlikely to affect therapeutic outcomes, or they may represent theoretical interactions documented in labeling but not consistently observed in clinical practice.

An example is the mild reduction in absorption of certain antibiotics when taken with antacids containing aluminum or magnesium. While the interaction is real, it can be managed simply by separating administration times by two hours, and the clinical impact of minor timing overlap is negligible for most patients.

Minor interactions present a design challenge for CDS systems. Including them in alerts contributes to fatigue without providing meaningful clinical value. Excluding them entirely means the information is unavailable when a clinician specifically wants to check. The recommended approach is to make minor interactions available through on-demand lookup but exclude them from proactive alerting workflows.

The unknown category applies when an interaction is documented in FDA labeling but the available evidence is insufficient to assign a meaningful severity level. This can occur when the label mentions a theoretical pharmacokinetic interaction based on in vitro data, when case reports exist but controlled studies do not, or when the interaction has been observed but its clinical significance has not been characterized.

Unknown severity does not mean the interaction is unimportant. It means the evidence base does not support confident classification. In practice, interactions classified as unknown should be surfaced to clinicians with appropriate context about why the severity is uncertain, allowing clinical judgment to fill the gap that evidence cannot.

RxLabelGuard uses the unknown classification conservatively. If label text describes a clear clinical risk with specific adverse outcomes, the interaction receives a severity rating even if the word 'severity' does not appear in the source text. The unknown category is reserved for genuinely ambiguous cases where the evidence does not point clearly to any severity tier.

RxLabelGuard's severity scoring operates in two passes against the FDA Structured Product Labeling text. The first pass is deterministic: a rule-based system scans for explicit severity language in the label. Terms like 'contraindicated,' 'do not use,' and 'must not be administered' map directly to the contraindicated level. Phrases like 'potentially fatal,' 'life-threatening,' and 'serious adverse reaction' indicate major severity. Monitoring language like 'dose adjustment may be required' or 'monitor closely' suggests moderate severity.

The second pass uses AI extraction via AWS Bedrock to handle the large volume of interaction descriptions that do not contain explicit severity keywords. FDA label text is often written as clinical narrative rather than structured data, describing mechanisms and outcomes without using standardized severity terminology. The AI pass analyzes the described mechanism, the potential adverse outcomes, and the recommended clinical actions to assign a severity level consistent with the deterministic rules.

Both passes are anchored to the source text. Every severity assignment includes the specific label excerpt that informed the classification, identified by spl_set_id and effective_time. This traceability allows consuming applications to present the evidence alongside the severity rating, enabling clinicians to evaluate the classification in context rather than treating it as an opaque score.

The practical value of severity classification is realized when it drives differentiated alert behavior in clinical decision support systems. A well-designed CDS implementation uses severity levels to determine three things: whether an alert fires, how the alert is presented, and what actions the alert requires.

For contraindicated interactions, the alert should be a hard stop requiring explicit override with documented clinical justification. The override reason should be captured in the patient record for audit purposes. For major interactions, the alert should be interruptive but overridable with a single acknowledgment click, and the alert content should include specific monitoring recommendations. For moderate interactions, the alert should be non-interruptive, appearing as an informational banner or sidebar notification that the clinician can review without workflow interruption. For minor interactions, the information should be available on demand but should not generate proactive alerts.

This tiered approach directly addresses alert fatigue. By reserving interruptive alerts for interactions that genuinely require immediate clinical attention, the system preserves the clinical significance of each alert. When a hard stop fires, the clinician knows it represents a contraindicated combination, not a minor timing suggestion, and responds accordingly.

Every interaction returned by the RxLabelGuard API includes a severity field with one of the five values: contraindicated, major, moderate, minor, or unknown. The response also includes the mechanism of interaction, a clinical recommendation, and the evidence snippet from the source FDA label that informed the severity assignment.

This structure allows consuming applications to implement their own alert policies on top of standardized severity data. A hospital EHR might configure hard stops for contraindicated and major interactions while suppressing minor interactions entirely. A pharmacy system might display all severity levels but use color coding and positioning to differentiate them visually. A clinical research platform might filter to only contraindicated pairs for adverse event monitoring.

The API does not impose alert behavior. It provides the severity classification and evidence needed for each consuming application to implement severity-appropriate responses within its own clinical workflow context.

This information is derived from FDA Structured Product Labeling and is provided for informational purposes only. It should not be used as a substitute for professional medical advice, diagnosis, or treatment. Always consult a qualified healthcare provider.