expert opinion

Optimizing design and analysis of clinical studies of nutrient effects

February 15, 2014


Robert P. Heaney, Ph.D., Osteoporosis Research Center, Creighton University, USA.

“The relationship between nutrient intake and the risk of harm due to deficiency or toxicity can be expressed in a U-shaped curve: the risk of deficiency is declining as intake increases to what would usually be considered as the adequate range of intake (Recommended Dietary Allowance); above this range (tolerable Upper Level of intake), pharmacologic or toxic effects begin to manifest, and thus the risk rises again. An inspection of the curve makes it clear that response to a change in nutrient intake will be dependent upon an individual’s starting nutritional status. If the basal status is deficient, then an increase in intake will usually produce a measurable benefit (or a reduction of risk). If the nutritional status is replete, an increase in intake will usually produce a null effect, and if the nutritional status is high, an increase in intake might be expected to increase risk of toxicity (or to decrease net benefit). This point is so obvious that one would think it would go without saying. Nevertheless, literally hundreds of studies of nutrient effects have reported null, or even adverse outcomes of nutrient interventions, almost always without reference to the basal status that prevailed in the study sample. Simi- larly, systematic reviews and meta-analyses have commonly pooled studies using different doses and different starting values, once again usually without reference to the relationship between nutrient dose and response.

As clinical nutrition studies generally focus on benefit, it is useful to concentrate on the left half of the dose- response relationship, with benefit rising with intake as a sigmoid-shaped curve: the extreme left end of the curve is often flat, i.e., increases in intake or dose produce little effect until intake reaches the ascending limb of the curve. Conversely, at the right end of the curve little effect is produced by yet higher intakes because the biochemical response systems have become saturated and are not capable of producing further effect when exposed to greater inputs. For most biochemical reactions and drug effects, the sigmoid curve extends over approximately three orders of magnitude (factors of 10), and since human dosing of test drugs is commonly focused on the mid-region of that curve, where the line is nearly straight, clinical responses to drugs are often treated as if they had linear characteristics. With nutrients, however, the entirety of the sigmoid is typically encompassed within an intake range spanning a single order of magnitude (e.g., the range of human calcium intakes extends from about 200 to 2,000 mg/day). If changing the intake of a nutrient is to have certain physiological effects, they will be found within those comparatively narrow ranges. A key consideration flowing from the sigmoid shape of the curve is the need to locate the intervention so that the basal status/intake is chosen so as to lie to the left of the ascending limb, and the change in intake is large enough to span much, or all, of the response region. However, several key studies have fallen into the trap of working mainly at either the low or the high end of the response curve, using nutrient doses which are too low to measure an effect or using actually adequate doses but for individuals who already had sufficient nutrient levels. As the transition from inadequate to adequate status occurs somewhere within the range of plausible intakes, studies intended to detect and quantify that effect must be centered on the intake range where the transition occurs. The response to a change in intake, for many, and perhaps most, nutri- ents, cannot safely be treated as if it were linear.

Given the reality of the U- and S-shaped nutrient dose-response curves, it seems useful to suggest several rules that should be kept in mind when designing (or interpreting) studies to evaluate specific nutrient effects:

1. Basal nutrient status must be measured, used as an inclusion criterion for entry into study, and recorded in the report of the trial;

2. The intervention (i.e., change in nutrient exposure or intake) must be large enough to change nutrient status and must be quantified by suitable analyses;

3. The change in nutrient status produced in those enrolled in the trials must be measured and recorded in the report of the trial;

4. The hypothesis to be tested must be that a change in nutrient status (not just a change in diet) produces the sought-for effect. As nutrient intakes (e.g., by supplementation) can be linked to problems of adherence, varying absorption and intrinsic biological differences in responsiveness, the focus needs to be on change in nutrient status (e.g., measuring blood nutrient concentrations at beginning and end of the study) as the independent variable;

5. Co-nutrient status must be optimized in order to ensure that the test nutrient is the only nutrition-related, limiting factor in the response. As nutrients interact with one another and the ability of an organism to respond to one is often dependent upon the status of several others (e.g., calcium effects usually need vitamin D for their expression and bone gain in response to calcium and vitamin D supplementation is dependent upon protein intake status), it is important that the nutritional status of the participants with respect to all related nutrients be optimized to study the effect of improving the status of the nutrient of interest.

It is puzzling and, indeed, surprising how often these rules are ignored or overlooked when studies are designed or their results are analyzed. However, the researchers state simply “nutrient X was without effect on system Y,” often with a finality suggesting that the case is closed. Only very few of the null studies that are commonly reported with respect to various nutrient effects state “change in status of nutrient X from level A to level B had no effect on outcome Y”. In addition, as basal values vary from individual to individual, resulting in different responses, averaging results across an intervention group will often blur, if not totally obscure, an underlying real effect. These considerations lead to a second set of guidance criteria, which is for the pooling of studies in systematic reviews and meta-analyses, now usually done primarily using methodo- logical rather than biological criteria:

1. The individual studies selected for review or meta-analysis must themselves have met the criteria for nutrient trials in the list above;

2. All included studies must have started from the same or similar basal nutrient status values;

3. All included studies must use the same or closely similar doses;

4. All included studies must have used the same chemical form of the nutrient and, if foods are used as the vehicle for the test nutrient, all studies must have employed the same food matrix;

5. All included studies must have the same co-nutrient status;

6. All included studies must have had approximately equal periods of exposure to the altered intake.

Basically, the criteria boil down to “pool like with like.” Rule 5, for example, is not followed in any systematic review or meta-analysis known to me. In any case, failure to observe these sets of rules, either in individual studies or in systematic reviews and meta-analyses, will inevitably bias the results toward the null. It may be objected that these rules are an ideal, and that we may not have the knowledge needed to apply them, even if we had the will. Nutrients are a heterogeneous lot. Not all the suggested rules have equal force for all nu- trients. It may be, for example, that co-nutrient status is less important for nutrient A than it is for nutrient B. Nevertheless, each rule reflects features that investigators need to consider and factor into their design and analysis. Additionally, the rules serve purposes beyond study design: 1) They allow us to understand why studies of actually efficacious agents might turn out to be null, especially if the rules had not been (or could not have been) followed. 2) In the case of systematic reviews and meta-analyses, the rules should stop us from continuing to cite studies as evidence of a certain conclusion when, in hindsight, we ought to have recognized that these studies could not validly have tested the associated hypotheses. 3) The rules help shape a research agenda, as they identify what it is we need to know in order to mount truly informative clinical trials of nutrient effects.”

Based on: Heaney R. P. Guidelines for optimizing design and analysis of clinical studies of nutrient effects. Nutrition Reviews. 2013; 72(1):48–54.