Valproate-Induced Hyperammonemia Despite a Therapeutic Level
A well-controlled epilepsy patient becomes acutely confused with a valproate level that hasn't changed and liver enzymes that are normal. The complication hiding behind those reassuring numbers has its own separate cause.
Teresa K., a 34-year-old high school librarian, ran nine miles on Sunday morning without difficulty, two months into training for her first half marathon. By Tuesday afternoon she was at the circulation desk repeating herself, unable to finish a sentence, and hunting without success for the words "due date" — a lapse her coworker of five years registered immediately, because it was so unlike her. She has had focal epilepsy since her mid-twenties, well controlled for the last five years on divalproex sodium extended-release, and hasn't had a breakthrough seizure in over three years. Her coworker, who has known her the whole five years, says this doesn't look like a seizure she recognizes — no convulsion, no post-ictal drowsiness resolving over twenty minutes, just a steady, strange confusion that hasn't budged in the hour since it started.
Her valproate level, drawn on arrival, comes back at 78 mcg/mL — comfortably inside the standard therapeutic range and identical, her outpatient chart shows, to her level from six months ago. Her liver enzymes are normal, which matters specifically because it rules out the more familiar valproate complication, hepatotoxicity, as the explanation for a sick-looking patient on this drug. What isn't normal is her ammonia, drawn on a hunch by the resident who remembered a lecture on exactly this: 112 µmol/L, several times the upper limit of normal, in a patient with no history of liver disease, no gastrointestinal bleed, and normal renal function to otherwise explain it. Valproate impairs ammonia clearance through a mechanism largely independent of its blood level or of standard liver injury markers: it depletes carnitine, a cofactor fatty-acid oxidation needs to keep the urea cycle adequately supplied, and it inhibits carbamoyl phosphate synthetase I directly, the urea cycle's own rate-limiting enzyme. That divergence — a textbook-normal valproate level and textbook-normal liver enzymes sitting next to a genuinely elevated ammonia and a genuinely confused patient — is exactly the signature of valproate-induced hyperammonemic encephalopathy, a complication that can occur even in a patient who has tolerated the same dose without issue for years. Far less settled is the supplement the team will argue about tonight. The clearest recommendation for L-carnitine comes from a pediatric neurology panel's consensus on childhood epilepsy — De Vivo and colleagues, Epilepsia, 1998 — which does name symptomatic valproate-associated hyperammonemia among its indications. But its groups are defined by young age, by polytherapy, or by an underlying metabolic disorder, and Teresa is thirty-four, on valproate alone, and ran nine miles the day before yesterday. That the mechanism is real is not the same evidence as that correcting it helps her.
On the epilepsy service, a level that isn't the problem
I want carnitine on board tonight, not pending a level that won't be back before the day is over. The mechanism here isn't speculative — valproate depletes carnitine, carnitine is what her mitochondria need to keep fatty-acid oxidation running, and when that pathway stalls, the urea cycle backs up and ammonia climbs, independent of her valproate level or her liver enzymes, both of which are exactly why this looks so deceptively reassuring on a standard panel. The case-series literature isn't randomized, I'll grant that upfront, but it's consistent across a real number of published cases — patients who get carnitine tend to clear both the ammonia and the confusion faster than reduction of valproate alone — and the De Vivo consensus names symptomatic valproate hyperammonemia as an indication in plain language, not as a hedge. She's confused right now. I don't want to be running a controlled experiment on how long that takes to resolve without it.
The case-series consistency you're describing comes overwhelmingly from a population that isn't her. The De Vivo consensus statement gets cited as though it settles this, and I'll grant you it names symptomatic hyperammonemia explicitly. But it is a pediatric neurology panel writing about childhood epilepsy, and most of the groups it singles out — young children, patients on multiple anticonvulsants stacking hepatotoxic and metabolic risk, patients with an underlying mitochondrial disorder, malnourished patients whose carnitine stores were already marginal before valproate touched them — describe a patient Teresa isn't. She's 34, on monotherapy, and training for a half marathon. Naming her complication in a pediatric document isn't the same as having studied her. What does reliably reverse this, in essentially every case series regardless of carnitine status, is reducing or stopping the valproate. I don't want us reaching for the unproven adjunct and quietly under-treating the one thing we know works.
"The mechanism isn't speculative" is true of the biochemistry — it isn't true of the claim that supplementing it changes her outcome specifically, which is a different, much less settled statement.
Both of you are arguing population-level priors about a question we can actually just measure. A free and total carnitine level exists, it won't change what happens in the next hour, but it will tell whoever's managing her tomorrow whether she's actually carnitine-depleted, rather than everyone continuing to argue from what population she resembles. In the meantime, reduce her valproate dose now — that's the one move neither of you is contesting, and it's the intervention with the actual outcomes literature behind it regardless of how the carnitine question resolves. Send the level, start carnitine tonight given how cheap and low-risk it is even without proof it's necessary in her specific case, and let tomorrow's result decide whether it continues past the acute encephalopathy or gets stopped once she's back to baseline.
Agreed: valproate dose reduced by roughly half rather than stopped outright, given her seizure history makes abrupt discontinuation a real risk of its own; L-carnitine started tonight despite the acknowledged evidence gap, on the strength of its low cost and low risk; a free and total carnitine level sent to settle, at least for her specifically, whether tomorrow's team is treating a confirmed deficiency or an empiric guess.
Not agreed: how long to continue the carnitine if her level comes back normal. The neurologist would continue it regardless, on the theory that a single serum measurement may not capture tissue-level depletion; the pharmacologist would stop it if the level is normal, on the theory that continuing an unproven therapy past the point evidence stops supporting even the empiric guess is its own kind of drift. That decision was left for tomorrow's team, with the level itself as the input neither side had this morning.