Medical Inspection – Health Blog

We examined the prescribing behavior of 3,646 physicians in relation to the introduction of 32 new drugs on the market from 1997 through 2000. The data reflected writing prescriptions for one drug per physician; the assignment of a particular drug to a physician was determined by  leters of the U.S. prescribing physicians, the study population was slighter older and more likely to be based in an office rather than in a hospital.

Physicians were enrolled from all 50 states, corresponding with the distribution of active physicians within the U.S. A Spearman Rank Order correlation of .932 between the number of physicians in the study and the number of physicians active within each state (as indicated in the American Medical Association’s annual survey of practicing physicians maintained by IMS Health, Inc.) indicated a strong rank order match between the two sets of data.

Dependent Variable

The drugs were indicated for the outpatient treatments of asthma and allergic rhinitis, hypertension, osteoarthritis and rheumatoid arthritis, depression, pneumonia, hypercholes-terolemia, and diabetes. The list of drugs appears in Table 1.

Table 1 Study Compounds Introduced into the Market from 1997 through 2000

Beclomethasone dipropionate (Qvar®, Ivax) Beclomethasone dipropionate inhaler (Vanceril®, Key) Budesonide (Rhinocort® Aqua, AstraZeneca)
Budesonide inhalation powder (Pulmicort® Turbuhaler, AstraZeneca) Budesonide inhalation suspension (Pulmicort® Respules, AstraZeneca) Candesartan drug cilexetil (Generic Atacand®, AstraZeneca) Celecoxib generic (Celebrex canadian, Pfizer) Cerivastatin (Baycol®, Bayer)* Cyclosporine tablet (Neoral generic, Novartis) Diclofenac canadian/misoprostol (Arthrotec® 75, Pfizer) Etodolac (Lodine® XL, Wyeth) Flesinoxan (Solvay) Glimepiride medication (Amaryl drug, Aventis) Insulin aspart (Novolog®, Novo Nordisk) Insulin lispro (Humalog®, Lilly) Drug Irbesartan (Avapro generic, Bristol-Myers Squibb) Leflunomide canadian (Arava generic, Aventis) Levofloxacin generic (Levaquin canadian, Ortho-McNeil) Loratidine (Claritin®, Schering) Meropenem (Merrem®, AstraZeneca)
Mibefradil (Posicor®, Roche)*
Mometasone (Nasonex® Nasal Spray, Schering) Mometasone furoate inhalation powder (Asmanex®, Schering-Plough) generic montelukast sodium (Singulair tablet, Merck) Moxifloxacin tablet (Avelox medication, Bayer) Naproxen canadian (Naprelan®, Elan) Olanzapine drug (Zyprexa generic, Lilly) Quinupristin/dalfopristin (Synercid®, Monarch) Repaglinide (Prandin®, Novo Nordisk) Rofecoxib (Vioxx®, Merck)* Sibutramine (Meridia®, Abbott) Sparfloxacin drug (Zagam generic, Bertek) Telmisartan drug (Micardis medication, Boehringer Ingelheim) Troglitazone (Rezulin®, Parke-Davis)* Drug Valsartan (Diovan canadian, Novartis) Generic Valsartan/HCT (Diovan drug HCT, Novartis) Venlafaxine generic (Effexor medication XR, Wyeth) Zafirlukast (Accolate®, AstraZeneca) Zileuton (Zyflo®, Abbott)

Although hospital pharmacies are a major source of prescription fulfillment, it is often difficult to link an individual prescription to a particular physician. Therefore, by concentrating on outpatient indications, we were able to better ensure the comprehensiveness of the prescribing data for each physician appearing in the study.

We originally designed the sample to test the relationship between participation in a clinical trial and subsequent prescribing of the study drug. The original analysis compared physicians who had participated in a clinical trial with a matched set of (control) physicians who had not participated in clinical trials of any sort in the previous five years. An analysis of both the clinical and the control physicians demonstrated that clinical trial physicians were more likely to prescribe a study drug after it had been on the market for a period of at least 18 months.

Approximately 50% of the study’s physicians had served as clinical investigators, and 50% constituted the matched control set. As with the general physician population in the U.S., most clinical trial sites are office-based, not hospital-based. One particular type of hospital—the major academic medical center—sometimes receives extensive press coverage for its work in clinical research. However, these centers constitute a decreasing proportion of all phase 3 clinical trial sites, and they perform a minority of all phase 3 studies. Most clinical investigators see patients in their office-based practice and enroll patients for clinical studies from these practices.

Table 2 Study Population and Physician Demographics in the U.S.

We conducted a comparative analysis (tests of statistical independence) between the two investigator and control groups of physicians and analyzed other variables that might have affected new drug-prescribing behavior. We found that although physicians who had worked as clinical investigators were more likely to prescribe the study drug when it arrived on the market, the relationship of the other demographic, practice, and prescribing variables in explaining new drug prescribing did not differ in any meaningful theoretical or statistical way. Hence, we decided to combine both sets of physicians into one data set, and, during subsequent analyses, to statistically control for the impact on new drug prescribing levels of a physician’s participation in at least one phase 3 clinical trial for the new drug. The analysis always tested for the appearance of statistical interaction between the two sets of physicians, the independent variables, and the likelihood of a physician’s being an early new drug adopter.
antibiotics online pharmacy

Study Population

We obtained the names of physicians from the IMS Health, Inc., database of active prescribing U.S. physicians (Table 2). Compared with known parameters of the U.S. prescribing physicians, the study population was slighter older and more likely to be based in an office rather than in a hospital.

Physicians were enrolled from all 50 states, corresponding with the distribution of active physicians within the U.S. A Spearman Rank Order correlation of .932 between the number of physicians in the study and the number of physicians active within each state (as indicated in the American Medical Association’s annual survey of practicing physicians maintained by IMS Health, Inc.) indicated a strong rank order match between the two sets of data.

Dependent Variable

The dependent variable is dichotomous. Physicians were characterized as either new drug adopters or not new drug adopters. The new drug adopters prescribed the new drug at some time during the first six months after its launch and continued to prescribe it for the next 12 months. We chose an initial timeline of six months (a commonly used industry convention). The date of the product launch was calculated as the date of the drug’s first prescription, as recorded by IMS Health, Inc.

All other physicians were characterized as not being new drug adopters. Physicians who prescribed the drug during the first six months—but who did not continue to prescribe the drug over the next 12 months—were not considered to be adopters of the new drug.
canadian cialis online

Table 3 Independent Variables in the Early Drug Adoption Logistic and Ordinary Least Squares Models

The term “early adopters” is derived from Rogers, who stated that “innovators” and “early adopters” made up 16% of individuals overall. In our study, the new adopters constituted 22% of the total number of physicians in the study.

Drugs are designated by their Uniform System of Classification (USC) code. IMS Health, Inc., and a majority of pharmaceutical manufacturers created the USC in 1975. The system uses five digits to standardize and categorize all pharmaceuticals in the U.S. on the basis of product type. USCs are used in the U.S. and Canada. In Europe, the equivalent classification is called an Anatomical Therapeutic Chemical (ATC).

USCs have four levels of hierarchy. USC2 is the broadest category, and USC5 is the most detailed category, allowing for more specificity within a category. The study used the USC5 level of specificity. For example:

• USC2:              respiratory therapy
• USC3:              bronchodilators, general
• USC4:              beta agonists
• USC5:              beta agonists, aerosol

New drug prescribing data were available for all the physicians in the study.

Independent Variables

The independent variables can be divided into several categories:

• physician demographics
• the physician’s practice and prescribing behavior
• the pharmaceutical marketing effort
• the drug’s therapeutic novelty

The data on the physician’s sex and age; whether the physician is based in a hospital or in an office; the physician’s specialty; and the physician’s board certification were derived from the AMA’s annual survey of practicing physicians. Apcalis Oral Jelly

The IMS Health database provides the information for the prescribing variables used as independent variables (Table 3):

• Total Pre-product Launch Prescribing Volume
• Total Pre-product Launch Drug Class Prescribing
• Pre-product Launch Company Prescribing Loyalty

Prescribing data were available only at the Total Volume level for these variables, not for the individual drugs within each of the prescribing categories. A drug was classified as either “first in class” or as “follow-on,” according to its respective order of appearance on the market in accordance with the IMS Health, Inc., USC coding scheme.

IMS Health, Inc., also provided the rank order of spending data used for the variable Pharmaceutical Marketing Support. Although this variable was not a physician demographic or a practice characteristic, we included it in a control function. The various drugs in our analysis came from companies in the largest revenue category to those with sales under $1 billion. The use of this variable helped to control for the role of differential marketing expenditures in understanding new drug adoption and thus helped to isolate the explanatory importance of physician demographic and practice characteristics.

We obtained information about a physician’s participation in a drug’s phase 3 clinical trials from a pharmaceutical industry database of clinical trials and the U.S. Food and Drug Administration’s (FDA’s) database of 1572 Forms, filed as part of new drug clinical trial activity.²⁰ Missing data never exceeded 2% of any variable.

Logistic Regression

The analysis employed binomial (binary) logistic regression, which is used when the dependent variable is categorical and the independent variables are either categorical or interval. Logistic regression is considered to have overcome many of the restrictive assumptions of ordinary least squares (OLS) regression. It does not require the assumption of a linear relationship between the independent and dependent variables, and it is not necessary for the dependent variable to be normally distributed. There is no assumption of homogeneity of variance, one does not need to assume normally distributed error terms, and the independent variables do not need to be unbounded.
buy generic cialis

Logistic regression applies a maximum likelihood estimation after transforming the dependent into a logit variable, or the natural log of the odds of the dependent occurring or not. Logistic regression estimates the probability of a certain event occurring in the dependent variable; it calculates changes in the log odds of the dependent variable, not the changes in the dependent variable itself, as OLS regression does. It is the likelihood, or probability, that the observed values of the dependent variable may be predicted from the observed values of the independent variables. The likelihood probability, like any probability, varies from zero (0) to one (1).

The log likelihood (LL) is its log, and it varies from 0 to minus infinity; it is negative because the log of any number less than 1 is negative. LL is calculated through iterations, making use of the maximum likelihood estimation (MLE).

The log-likelihood test of a model can be used to estimate the statistical significance of the entire model. Frequently called the “model chi-square test” or the “likelihood ratio test,” it is based on -2LL (deviance). It is an alternative to the Wald statistic. The model chi-square provides the most frequently used significance test for a logistic model. A well-fitting model has a P value (Sig.) of .05 or lower; that is, we want model chi-square to be significant at the .05 level or better.

The significance of individual parameters can also be estimated. Logit coefficients, also called unstandardized logistic regression coefficients, are similar in interpretation to the B (unstandardized regression) coefficients in OLS regression. Logits are simply the natural log of the odds; they are used in a logistic regression equation to estimate the log odds that the dependent equals 1 in a binomial logistic regression.

Partial R is a method of assessing the relative importance of the independent variables, similar to beta weights or standardized partial regression coefficients in OLS regression.

The odds ratio is another method of determining the relative importance of the independent variables. It avoids some of the interpretative difficulties involved with the Wald statistic, particularly the greater possibility of type II errors. In this study, the odds ratio is used to help readers understand the relative importance of each independent variable.
buy cialis soft tabs

Hosmer and Lemeshow and Menard present a more extensive discussion of logistic regression.

Related Posts

• Demographics, Practices, and Prescribing Characteristics: DISCUSSION
• Demographics, Practices, and Prescribing Characteristics: RESULTS
• Demographics, Practices, and Prescribing Characteristics