Plant Protein vs. Processing: What the Heart Data Show

Swapping cow's milk for soymilk sounds like a straightforward heart-health move. Many dietary guidelines endorse it. The complication: commercial soymilk can be ultra-processed, and sweetened versions add sugar. That puts one food in two categories that health advice often treats as opposites.

Current evidence does not settle whether soymilk's processing classification changes its cardiometabolic effects. But a small randomized feeding trial offers a useful clue on the larger question: when people eat minimally processed diets, does choosing lentils instead of pork change insulin resistance, cholesterol, inflammation, or body fat? In that trial, both approaches improved those markers to a similar extent. The result points toward overall dietary quality and processing level as major variables, but it does not prove that ultra-processed food was the sole cause of the change.

The soymilk question remains open

A 2024 systematic review and meta-analysis by Erlich et al. examined randomized trials that replaced cow's milk with soymilk. Its protocol included a pre-registered comparison of sweetened and unsweetened soymilk for LDL cholesterol, non-HDL cholesterol, blood pressure, and C-reactive protein (CRP).

That design addresses the question people actually face in the grocery aisle: does added sugar change the health tradeoff when soy replaces dairy? The accessible summary of the review does not provide the pooled effect sizes, confidence intervals, heterogeneity measures, certainty ratings, or subgroup results. So it cannot support a numerical claim that sweetened or unsweetened soymilk improves—or worsens—any of those outcomes relative to cow's milk.

The practical reading is modest. Unsweetened soymilk is a reasonable default if you want to avoid added sugar. But NOVA classification alone cannot tell you whether a specific fortified soy beverage will produce a better lipid or blood-pressure result than dairy. A category label is not a clinical trial.

What happened when pork and lentils were both minimally processed?

The most direct controlled evidence comes from PRODMED2, a 2025 randomized crossover feeding trial by Vaezi et al. The study enrolled 36 community-dwelling older adults. Each participant followed two diets: a minimally processed pork-centered pattern and a minimally processed lentil-centered lacto-ovo-vegetarian pattern.

Both diets cut ultra-processed food from roughly 50% of energy at baseline to about 13% during the intervention. Compared with baseline, each diet significantly improved HOMA-IR, fasting insulin, C-peptide, total cholesterol, LDL cholesterol, apolipoprotein B, non-HDL cholesterol, and CRP; the reported p values were all below 0.038. The study found no statistically significant between-diet difference for any of those endpoints.

Weight changed in the same direction. Participants lost 3.8 kg on the pork-centered diet and 4.4 kg on the lentil-centered diet. Fat mass fell by 2.6 kg and 2.9 kg, respectively. Neither between-diet comparison was statistically significant. Fasting leptin declined by 1.9 ng/mL on the pork diet and 2.5 ng/mL on the lentil diet, while FGF21 rose by 65 pg/mL and 88 pg/mL. Again, both changed significantly from baseline and did not differ significantly between diets.

Those are biomarker and body-composition findings, not cardiovascular event data. The trial did not show fewer heart attacks, strokes, or deaths. It showed that, in this specific group over the study periods, two low-UPF dietary patterns produced similar changes in risk markers.

Why the trial cannot crown UPF the winner

The tempting conclusion is that processing mattered and protein source did not. The data are consistent with that explanation, but the study cannot isolate it.

Participants spontaneously ate 333 to 437 fewer calories per day during the two interventions. They also lost substantial weight. Calorie reduction and weight loss can improve insulin resistance, lipids, inflammation, leptin, and related markers without any UPF-specific mechanism. Because the trial did not hold calories steady while changing processing level, it cannot separate the effects of reduced UPF intake from the effects of eating less and losing weight.

The null comparisons also do not establish equivalence between lentils and pork. With only 36 participants, the study had limited power to detect modest differences. The reported analysis did not include formal equivalence testing or pre-specified clinical equivalence margins. "No statistically significant difference" means the study did not detect a difference; it does not mean the diets have identical effects.

The crossover design adds another uncertainty. The two diet periods had a two-week washout. Lipid markers such as LDL cholesterol and apoB can respond to diet over weeks, so effects from the first intervention could have carried into the second. That could blur a real difference in either direction.

The follow-up supports reversibility, not a mechanism

At one-year follow-up, participants' UPF intake had risen to about 44% of energy, and measured biomarkers moved back toward baseline. That pattern suggests the improvements did not persist after participants partly returned to their prior dietary pattern.

It does not prove that UPF intake alone drove the reversal. The follow-up was not a new randomized intervention, and intake returned to about 44% of energy rather than the original 50%. The result supports a relationship between abandoning the intervention pattern and losing its biomarker gains. It cannot define a precise dose-response curve or show whether 44% and 50% UPF intake have meaningfully different effects.

Diet quality has more than one dimension

Evidence from broader dietary-pattern research argues against treating processing as the only meaningful feature of a diet. A 2024 network meta-analysis by Bonekamp et al. in people with established cardiovascular disease found that dietary patterns differed in their effects on cardiovascular risk factors. Its findings support the idea that macronutrient composition and food-group quality matter alongside processing level.

That matters for interpreting PRODMED2. The pork and lentil diets may have converged because both improved several features at once: minimally processed foods, food quality, macronutrient balance, and likely fiber and saturated-fat profiles. The study did not manipulate each feature separately, so it cannot assign the gains to one variable.

A separate 2024 network meta-analysis of randomized dietary interventions likewise found that dietary patterns can differ in cardiovascular outcomes. It does not answer whether UPF classification or plant-versus-animal protein explains those differences, but it reinforces a less tidy conclusion: diet quality is not one dial.

The case for vegan diets as a distinct cardiovascular intervention is also less certain than popular summaries imply. A Cochrane review by Rees et al. found limited and highly uncertain evidence for vegan dietary patterns in primary and secondary cardiovascular prevention. Many included studies were short, small, or at moderate-to-high risk of bias. That evidence does not show that vegan diets fail; it means the trials have not established a reliable cardiovascular benefit independent of the other changes people often make when they adopt them.

What to do with this evidence

If your diet gets about half of its calories from ultra-processed foods, replacing a large share with minimally processed meals is a sensible experiment. PRODMED2 found meaningful improvements in metabolic and lipid markers while participants moved from about 50% to 13% UPF energy, regardless of whether their main protein was pork or lentils. The same study cannot tell you how much reduction is enough, whether smaller changes work, or whether the results apply to younger adults, people with established cardiovascular disease, or metabolically healthy people.

Choose protein based on the whole meal and your preferences, then inspect the surrounding diet. Lentils bring fiber and fit easily into minimally processed meals. Minimally processed animal foods can also fit a high-quality pattern. Neither fact turns protein source into the main lever shown by this small trial.

The evidence favors improving the pattern first: fewer ultra-processed foods, better food-group quality, and a diet you can sustain. Soy versus dairy and lentils versus pork remain narrower questions than the health marketing around them suggests.

For soymilk specifically, unsweetened options remain the cautious choice when added sugar is avoidable. The relevant meta-analysis asked the right question, but the available results do not yet justify a confident claim about the size—or even the direction—of the sweetened-versus-unsweetened difference.