An atlas of drug-nutrient interactions at the scale of a national pharmacopoeia: a protocol for classifying 664 molecules into three supplementation outcomes

Abstract

Context. Nutritional depletion associated with long-term treatments is described in landmark reviews, yet it remains poorly codified in clinical practice and is the subject of no systematic reference base accessible to the public. Objective. To build and publish an atlas linking every medication marketed in France to an explicit conclusion on whether supplementation is warranted, according to reproducible and verifiable criteria. Method. The 8,193 product names in the ANSM registry were normalized then reduced to 664 molecules identified by their International Nonproprietary Name. For each one, the literature (PubMed and PubMed Central, systematic reviews, positions of authorities: FDA, NIH ODS, EFSA) was examined through a cascade of criteria: mechanistic plausibility, depletion measured in humans under treatment, magnitude and dose or duration relationship, clinical relevance, quality of evidence, existence of a documented corrective form and dose. Each molecule received a single outcome: internal reference, third-party option (active ingredient and form indicated, with no commercial link) or nothing to add. Results. Of 664 molecules, 31 (4.7%) fall under an internal reference, 59 (8.9%) under a third-party option and 574 (86.4%) call for no supplementation. Scope. The negative conclusion dominates very widely; the boundary between the two positive outcomes is commercial and not evidentiary, which the article sets out and submits for discussion.

Introduction

Prolonged use of a medication may be accompanied by a gradual decline in the status of one or more vitamins, minerals or trace elements. This drug-associated nutritional depletion, often subclinical before it becomes measurable, was brought to attention by the synthesis of Samaras et al., explicitly subtitled “a story rarely told” ¹, then updated therapeutic class by therapeutic class by Mohn et al. ². Three main pathways are described: reduced absorption (proton pump inhibitors and the release of dietary vitamin B12, iron or calcium), disrupted metabolism (enzyme inducers and vitamin D, statins and endogenous synthesis of coenzyme Q10 through inhibition of the mevalonate pathway) and increased excretion (diuretics and magnesium, potassium, thiamine or zinc) ².

Unlike drug-drug interactions, these drug-nutrient interactions are not integrated into clinical routine and remain rarely assessed in everyday practice ^2,3. Péter et al. underline their public health relevance, particularly in older, polymedicated people ³. Some signals are nonetheless taken up by authorities: the professional fact sheet of the NIH Office of Dietary Supplements states that metformin may reduce the absorption of vitamin B12 ⁴, and in 2011 the FDA issued a safety communication linking prolonged use of proton pump inhibitors to cases of hypomagnesemia ⁵.

There is, however, to our knowledge, no systematic, publicly accessible reference base that starts from the pharmacopoeia as it is actually prescribed and assigns each medication an explicit conclusion, including a negative one. Existing reviews proceed by selected classes and leave the reader without an answer for the great majority of treatments; conversely, the commercial literature of dietary supplements tends to generalize isolated signals well beyond what the data allow. The aim of the present work is to describe the construction of such a reference base, the Layer Nutrition atlas: exhaustive coverage of a national registry, an explicit classification protocol into three mutually exclusive outcomes, publication of the resulting distribution and of its grey areas, in order to expose the method to criticism.

Materials and methods

Starting point. The entry point is the public registry of pharmaceutical products of the ANSM, that is 8,193 brand and generic names. These names were normalized by deduplicating dosages, dosage forms and combinations, then reduced to 664 molecules identified by their International Nonproprietary Name (INN). Each entry in the classification file carries the INN, the pharmacological class, the use, the family, the assigned outcome and, where applicable, the internal reference concerned or the list of orientation active ingredients. The scope covers long-term treatments and short treatments with repeated prescription, since the question of depletion arises essentially over exposures of several months ^1,2.

Literature search strategy. For each molecule or homogeneous group of molecules, the search combined three tiers. The first consists of landmark cross-sectional reviews of drug-nutrient interactions ^1,2,3, used as an initial mapping of candidate drug-nutrient pairs. The second brings together the systematic reviews and meta-analyses specific to each class where they exist, for example on proton pump inhibitors in older people ⁶. The third gathers health authority documents: FDA safety communications ⁵, NIH ODS professional fact sheets ⁴, EFSA methodological guidance ⁷. Searches were conducted on PubMed and PubMed Central from the INN crossed with the candidate nutrients identified at the mechanistic tier. Only published and verifiable sources (PMID or DOI identifiers, official authority pages) were retained; no unsourced data were admitted into the file.

Evidence inclusion criteria. The studies retained had to concern humans under treatment; animal or in vitro data served only to establish mechanistic plausibility. Weighting followed the classic hierarchy of study designs, meta-analyses and randomized trials at the top, then cohort studies, then case series and reports ⁸, while taking into account the limitations specific to nutrition research, where the randomized trial runs up against the absence of a strict dietary placebo, imperfect blinding and variable adherence, which preserves a complementary value for observational studies ^8,9. Nutritional status was assessed by distinguishing markers of exposure, status, function and effect, following the framework of the BOND program ¹⁰, and referred to the reference intakes (EAR, RDA, AI, UL) defined by the Food and Nutrition Board and disseminated by the NIH ODS ¹¹. Finally, the register of conclusions was bounded by the standards for substantiating claims: EFSA requirements on the characterization of the constituent and the demonstration of a measurable physiological effect in humans in the target population ⁷, and the gradation of claims defined by the FDA, from Significant Scientific Agreement to structure or function claims ¹².

Classification protocol. Each molecule was evaluated through a cascade of five criteria, applied in order and documented: (i) mechanistic plausibility, an identified pathway (absorption, metabolism, excretion) linking the medication to a specific nutrient; (ii) depletion measured in humans under treatment, at least one study measuring a decline in a status marker in exposed people, a purely mechanistic deduction being insufficient; (iii) magnitude and clinical relevance, assessed by the size of the decline, its relationship with dose or duration, the existence of a functional impact or of identified at-risk populations, ideally objectified by a functional marker; (iv) quality of evidence, the position of the available studies in the hierarchy of study designs and the convergence of their results; (v) documented correction, the existence of a form and an intake dose whose corrective effect on the marker is described.

The assignment rule is conservative. The default outcome is “nothing to add”: it is retained whenever criterion (ii) is not satisfied, or when the magnitude and clinical relevance (iii) are not documented in a convergent way. When the chain (i) to (v) is sufficiently complete, the molecule receives a positive outcome, which always designates a single-compound active ingredient and its form, never a complex. The split between the two positive outcomes does not rest on evidence: a molecule is classified “our reference” if its class corresponds to one of the five formulations in the catalogue (lipid-lowering drugs, metformin, proton pump inhibitors, oral contraception, beta-blockers), and “third-party option” otherwise, the atlas then indicating the active ingredient and form to consider, with no link and no seller. This commercial character of the last split is acknowledged and discussed below. The formulation of the conclusions remains strictly associative: the atlas reports that studies measure or associate a decline in a marker with a treatment, and never asserts that a medication causes a deficiency. An internal audit review was then carried out on all the classifications; its observations, including unfavourable ones, feed the Discussion and Limitations sections.

Results

The 664 molecules break down as follows: 574 (86.4%) as “nothing to add”, 59 (8.9%) as “third-party option” and 31 (4.7%) as “our reference”. The first result is therefore the massive dominance of the negative conclusion: for nearly nine molecules out of ten, the literature does not assemble the chain of evidence that would justify an intake. This parsimony is not an editorial choice but the direct consequence of the protocol, the negative outcome being the default outcome of the cascade.

The 31 molecules of the “our reference” outcome are distributed across five classes: ten beta-blockers, nine oral contraception molecules (ethinylestradiol and eight progestins), six lipid-lowering drugs (five statins and fenofibrate), five proton pump inhibitors and metformin. The signals that carry these classes are of uneven strength. The most robust concerns metformin: a systematic review including randomized trials measures a significant decline in serum vitamin B12 under prolonged treatment ¹³, the reported prevalence of the deficiency varies, across studies, on the order of 6 to 50% with a risk increasing with dose and duration ¹⁴, and the NIH ODS professional fact sheet retains this interaction ⁴. For proton pump inhibitors, the FDA safety communication documents the hypomagnesemia signal under prolonged use ⁵, while a recent systematic review in older people measures above all a decline in vitamin B12, its results on magnesium remaining inconsistent ⁶. For statins, studies measure a decline in circulating coenzyme Q10, but the intramuscular data and the value of supplementation remain inconsistent ¹⁵. For oral contraceptives, the literature associates their use with a lower vitamin B6 status, the effect on folates and vitamin B12 being more debated at current doses ¹⁶.

The 59 molecules of the “third-party option” outcome form four large sets. The most numerous brings together 42 antibiotics and antituberculosis agents oriented toward probiotics. Eight molecules of the bone field (bisphosphonates, denosumab, raloxifene, teriparatide, oral corticosteroids) are oriented toward calcium and vitamin D, consistent with the recommendation of the American College of Rheumatology for systemic corticosteroids ¹⁷. Five diuretics (thiazides and furosemide) are oriented toward magnesium and potassium. There remain singular cases with strong evidence: methotrexate toward folates, isoniazid toward vitamin B6, and two enzyme-inducing antiepileptics (carbamazepine, phenobarbital) toward vitamin D.

The “nothing to add” outcome brings together 574 molecules. It covers both cases where mechanistic plausibility is lacking and cases where a hypothesis exists without convergent human measurement. The protocol produced coherent internal distinctions there, which constitute so many qualitative controls: non-inducing antiepileptics (levetiracetam, lamotrigine, gabapentin, pregabalin, lacosamide, topiramate, zonisamide) are kept there while enzyme inducers are oriented toward vitamin D; inhaled, nasal and topical corticosteroids are kept there on the grounds of low systemic exposure, while their oral counterparts fall under a third-party option; menopausal estradiol is kept there while contraceptive ethinylestradiol, more highly dosed, falls under a positive outcome.

Discussion

The distribution obtained is first of all a result of method: applied with demanding criteria, the question of whether something should be added receives a negative answer in 86.4% of cases. This figure is consistent with the state of the literature, where landmark reviews underline the predominance of small observational studies, the heterogeneity of biomarkers and the absence of quality intervention trials for most of the associations described ^1,2. It also matches the severity of the substantiation standards: the demonstration required by EFSA for a health claim is reached by only a minority of dossiers ⁷, and the claims authorized by the FDA under the Significant Scientific Agreement standard remain rare ¹². A reference base that recommended an intake for the majority of medications would, conversely, be a methodological warning sign.

The critical examination of the boundary between “our reference” and “third-party option” constitutes, in our view, the most useful contribution of the exercise. This boundary is commercial, not evidentiary, and it would be misleading to present it otherwise: several third-party options rest on evidence at least as solid as that of the internal references. The co-prescription of folates with methotrexate and of vitamin B6 with isoniazid is standard of care; the intake of calcium and vitamin D accompanying bisphosphonates and denosumab is a safety requirement; the ACR recommendation on systemic corticosteroids is explicit ¹⁷. Conversely, the most fragile class among the internal references is that of beta-blockers: the historical signal of a decline in coenzyme Q10 concerns above all non-selective propranolol, and the reviews devoted to coenzyme Q10 in cardiovascular disease, centred on heart failure and statins, do not show any established depletion under cardioselective beta-blockers ^15,18. On the sole evidence of depletion, this class would probably not cross the threshold of a positive outcome; its classification reflects the existence of a formulation in the catalogue, which the present publication documents rather than conceals.

The largest contingent of the “third-party option” outcome, 42 anti-infectives oriented toward probiotics, stems from a shift of model that must be named: it is not a nutritional depletion in the sense of the previous sections, but support of the microbiota. The evidence there is of moderate magnitude, documented for the prevention of antibiotic-associated diarrhea, above all in children and on systemic courses ¹⁹, and conditional on the baseline risk for the prevention of Clostridioides difficile diarrhea, with safety reservations in frail or immunocompromised people ²⁰. The extension of this orientation to single-dose exposure, such as fosfomycin in cystitis, or to hospital anti-infectives administered intravenously, is contextually questionable; the borderline case is oral vancomycin, which treats precisely C. difficile colitis and for which an “add-on” orientation invites confusion.

Limitations

The internal audit of the classifications, complemented by an adversarial review, identified inconsistencies and debatable cases, which we report rather than smooth over. The clearest concerns dexamethasone: a systemic glucocorticoid of the same register as prednisone and prednisolone, it is classified “nothing to add” whereas the latter fall under an orientation toward calcium and vitamin D; the ACR recommendation, which targets all systemic glucocorticoids, dexamethasone included, argues for alignment ¹⁷. Five candidacies for a third-party option remain under discussion. Levodopa, whose prolonged exposure is associated, in comparative work, with a decline in B vitamins and with peripheral neuropathies ²¹. Valproate, for which a secondary carnitine depletion is described, with evidence of supplementation efficacy judged mixed outside at-risk populations ²². Captopril, whose thiol group is associated with increased urinary excretion of zinc not found with enalapril, a weak signal not generalizable to the other angiotensin-converting enzyme inhibitors ²³. Finally the calcineurin inhibitors, tacrolimus and ciclosporin, whose use is associated in humans with renal magnesium loss and frequent hypomagnesemia, reversible by a magnesium intake ²⁴ : the chain from mechanism to documented correction is complete there, and their retention in “nothing to add” illustrates that the depletions retained by the atlas lean toward the classes where an offer exists, since these are precisely third-party options with no product in the catalogue. Conversely, the probiotic orientation of single-dose fosfomycin, like that of vancomycin whose oral form treats precisely Clostridioides difficile colitis, appears to exceed or contradict what the evidence allows ^19,20.

Other limitations relate to scope and method. The rule of a single outcome per molecule flattens secondary drug-nutrient pairs: oriented toward probiotics on account of its effect on the microbiota, cotrimoxazole thus sees the antifolate character of trimethoprim recede into the background, and the diuretics their second candidate nutrient. In the same spirit, the field of orientation active ingredients would gain from being refined without a change of outcome: furosemide could mention thiamine, whose urinary loss under loop diuretics is described in landmark reviews ², and the thiazides zinc. The scope, set on the normalized ANSM registry, also leaves outside the atlas molecules with documented depletion but little or no longer prescribed in France, such as cholestyramine and fat-soluble vitamins, phenytoin, sulfasalazine, orlistat or H2 antihistamines ^1,2 : these are not classification errors, but a coverage bounded by the choice of source registry; since most of these molecules would fall under a third-party option with no product in the catalogue, their absence tends to inflate the share of “nothing to add”, and the figure of 86.4% must be read in the light of this scope. The classification was moreover carried out by a single operator from published sources, with no external review committee to date; we claim no professional validation that did not take place. Finally, the raw material is mostly observational, with heterogeneous biomarkers, dosages and durations, and the limitations specific to the randomized trial in nutrition ⁹ make certain conclusions sensitive to the interpretive frame retained; the outcomes must therefore be read as dated judgments, revisable as the literature evolves.

Conclusion and perspectives

The atlas shows that an exhaustive reference base of drug-nutrient interactions, bounded by explicit criteria, leads massively to a negative conclusion, and that this parsimony is the price of rigour. The method is reproducible: the starting base is public, the criteria of the cascade are stated, the sources are identifiable by their PMID, DOI or official pages, and the classification file is structured to be audited line by line.

The perspectives are of three kinds. First, periodic updating, at the pace of revisions of the ANSM registry and of the publication of systematic reviews by class, of which the recent example of proton pump inhibitors in older people ⁶ illustrates the contribution. Next, the resolution of the debatable cases listed under Limitations, starting with the alignment of systemic corticosteroids and with the clarification of the status of probiotic orientations, which fall under a model distinct from nutritional depletion. Finally, the opening of the classification to external criticism: researchers, pharmacists and clinicians are invited to contest an outcome by producing verifiable sources, the burden of proof holding in both directions. The distinction between what the literature measures and what a catalogue sells, made explicit here down to the definition of the outcomes, seems to us to be the condition for an object built by a brand to be read, cited and discussed as an object of method.