The JNC VI bases its recommendations on the drug treatment of hypertensive people on a convincing scientific basis. The pooled results of the trials on diuretics and/or β-blockers considered as the basis for the guidelines show a clear reduction in stroke, [22, 46] congestive HF and total cardiovascular mortality.[22, 23, 46] Nonetheless, differences between the populations studied and the populations actually receiving the medications pose the problem of the generalizability of the results. Using only few eligibility/exclusion criteria, we found that the generalizability of the individual trials was poor. Forty percent of the population would not have been enrolled in at least one trial, and half of the population was not eligible for at least two trials.
The JNC VI extends its recommendation to people with blood pressure levels generally lower than the population included in the trials. In one of the groups for which treatment is recommended by the JNC VI (i.e. people with high normal blood pressure and target-organ disease or diabetes), we found nobody that would be included in at least two trials. This is problematic. Since characteristics of the population translate into different baseline risks, the absolute benefits observed in trials may be different than those observed in the general population. While it is arguable that clinical trials, excluding people with substantial comorbidity, underestimate the real benefit of the treatment,  any generalization to a population with a baseline risk level different the one studied in an individual trial needs to be evaluated. Furthermore, since the risk of adverse drug reactions increases with the underlying clinical severity,  it is seldom easy to estimate the real risk/benefit ratio when data are not available in a specific population.
Our data show that a substantial proportion of people (30%) with a diagnosis of hypertension and that should receive treatment according to the JNC VI guidelines have not been prescribed any antihypertensive drug. Our results are in line with another report from NHANES III showing that about 50% of all people with a diagnosis of hypertension (regardless of the actual blood pressure) were not receiving treatment.
The decision not to treat might be a due to poor familiarity with the JNC guidelines,  to a risk/benefit evaluation tailored to the individual patient, or to patient preference. Our data show that the risk profile is much more important in influencing treatment than the actual blood pressure levels. In fact, people with stage 2/3 hypertension but without risk factors or target-organ disease were treated less frequently than people with lower blood pressure with additional risk factors or target-organ diseases. The higher prevalence of treatment in people at higher risk (advanced age, cardiovascular diseases, and diabetes mellitus) suggests that the characteristics of the individual patient are a major determinant of the decision to treat. Although this finding might be an artefact due to the fact that people receiving treatment for hypertension are likely to have normal blood pressure levels and therefore not captured by our study, other data from different settings support this interpretation. A community-based study in a older population showed that hypertensive people with additional risk factors were more likely to receive antihypertensive medications,  while a prospective study in a family medicine setting found that treatment for hypertension was independent from the actual blood pressure readings. Concomitant cardiovascular diseases were associated with increased use of antihypertensive in a nursing home population. The explanation that physician's own clinical judgment might override guidelines' recommendation is also in accordance with data on compliance with guidelines on diabetes care.
We found that, among people with a prescription and taking an antihypertensive drug at the time of the interview, the type of medication was not following the indication of the JNC VI. This failure of the JNC guidelines in influencing the type of therapy described has already been described in a study examining the effect of the fifth report of the JNC. However, it must be noted that this is a cross-sectional study and we have no information on which drugs have previously been tried.
Our data show no association between the generalizability of the trials and the decision to prescribe an antihypertensive medication. This study is underpowered to rule out an association, but the comparison of the point estimates of the OR of being treated in the crude analysis and the adjusted model confirms the role of the clinical characteristics in physicians' adherence. In the crude analysis, where this baseline risk is not taken into account, people to whom the clinical trials are not generalizable are more likely to be treated. In the multivariable analysis taking into account the baseline risk, the direction of the association is reversed. The reason for this finding probably lies in the fact that clinical trials considered tend to exclude high risk people (see table 2), who are the most likely benefit from treatment and therefore to be treated in clinical practice.
Translating trial findings into clinical decision making is never an easy task, and generalizability is one of the factors that should be taken into account when trying to implement "evidence based" interventions. Other related factors are also likely to play a role. Patient's preference is one of these, and has more far-reaching implications than just treatment choice. The prescription of a drug that the patient deems effective may have a beneficial effect in and of itself, something that has been defined "therapeutic effect of patient preference". While controlled clinical trials can avoid this problem by blinding the patients to the treatment they are receiving, in the clinical practice setting this is impossible. Furthermore, patients with strong preference are likely to refuse randomization, and this clearly affects generalizability. One possible solution is to use information coming from settings more similar to the clinical practice when making recommendations. For example, beside observational studies, valuable information can come from "pragmatic trials", that are performed in the real clinical practice, and aim to inform choices between treatments rather than to measure the benefit of a treatment under ideal circumstances.
The results of this study should be evaluated taking into account the limitations of our approach. First, people receiving an antihypertensive medication are more likely to have normal blood pressure, therefore were not in our sample. Although the proportion of people whose high blood pressure was controlled by anti-hypertensive drugs was relatively low (10%), this bias potentially leads to an underestimation of the adherence to guidelines. However, the treatment rate in this population (70%) is similar to that reported in another study in a different setting, in which the prevalence of treatment among people with known hypertension was 80%.
The diagnosis of hypertension and the decision on drug treatment were made before the blood pressure levels used in this study were measured. Therefore, the blood pressure at the time of the last physician's visit was not necessarily as high as at the moment of the study interview. The result would be an underestimation of the adherence rate. Although this bias cannot be discounted, the rate of treatment that we observed is similar to that observed in other studies.[61, 62] Furthermore, blood pressure was measured at a single point in time, and there is the possibility that a regression to the mean effect have reduced the number of patients appropriate for the JNC VI guidelines. Another source of bias might come from people who were prescribed lifestyle modification, and whose blood pressure had increased to level requiring drug treatment at the time of the interview. Again, we would have underestimated the adherence rate.
We control only for only a few eligibility/exclusion criteria. It is probable that the proportion of people that would have been enrolled in the clinical trials we considered is actually much lower than we report. For example, in the SHEP trial only 2.7% of people screened were eligible to the first baseline visit. Of these, less than half was actually randomized. If people to whom the trials are generalizable were actually more likely to receive treatment, this misclassification would result in a spurious negative association between generalizability and treatment. On the contrary, if people excluded from the trial are more likely to receive drug treatment, we might underestimate the negative association between generalizability and adherence to guideline. Data from the literature show that the second scenario is more likely, with people with risk factors that lead to exclusion from the clinical trial being more likely to be treated.
Finally, the precision of our estimates was quite low, and the inferences that can be made on the basis of our data are limited by this lack of statistical power.