The US Institute of Medicine concludes that “the reasons for defining modalities as ‘CAM therapies’ are not only scientific but also political, social, and conceptual”¹³ and notes that “Given the lack of a consistent definition of CAM, some have tried to bring clarity to the situation by proposing classification systems that can be used to organize the field.”¹⁴ Researchers classify therapies such as acupuncture, massage therapy, and chiropractic care as alternative medicine to ensure a sample that includes sufficient numbers of CAM users to allow for demographic statistical analysis. The inclusion of such therapies in a taxonomy of CAM is unfortunate, however, as the objective of behavioural science studies of CAM is to understand why people use therapies that are not accepted by physicians as being effective.¹² In other words, a central policy and practice issue is to understand why patients use Eisenberg’s “more alternative” practices. In order to understand why patients use unproven therapies against their physicians’ advice, it is necessary to first discern which CAM practices physicians are most likely to question in terms of effectiveness (ie, which are the “more alternative” practices).

The label alternative medicine as applied to practices such as chiropractic care, acupuncture, and massage therapy is a result of health claims about the therapies made by some CAM practitioners and not physicians’ judgments about the therapies as typically used.¹² Consequently, this research uses data from a survey of Alberta family physicians to suggest a taxonomy of CAM practices based on physicians’ assessments of the effectiveness of various CAM therapies.

Internationally, there has been much research on physicians’ perceptions of CAM.^15–18 However, whether this information can be generalized to Canada is questionable.¹⁹ In Canada, such work is of limited scope. As of 2008, there have been 6 published quantitative Canadian studies that have sought to address the issue of physicians’ assessment of and attitudes toward CAM.^19–24 Despite Astin and colleagues’ call for studies with larger samples¹⁵ (after analysis of 19 studies from around the globe), to date no such study in the Canadian context has been published. In none of the Canadian studies did the sample size exceed 500, which is the minimum number of cases required for robust multivariate analysis.²⁵ Further, the Canadian studies focus on a limited number of the more popular therapies. Goldszmidt and colleagues’ study of Quebec general practitioners considered chiropractic care, acupuncture, and hypnosis,²⁰ while a study of general practitioners in Alberta and Ontario by Verhoef and Sutherland that same year looked at a wider, but still limited, range of practices—chiropractic care, acupuncture, hypnosis, faith healing, osteopathy, homeopathy, herbal medicine, reflexology, and naturopathy.¹⁹ More recently, Kaczorowski and colleagues’ study of family physicians and specialists practising in Hamilton, Ont, looked at chiropractic care, acupuncture, homeopathy, herbal medicine, and naturopathy.²³ Focusing solely on the more popular practices obfuscates a central policy and practice issue regarding usage of CAM: Why would patients use a therapeutic practice if its effectiveness is rejected by physicians?

In order to answer this question it is useful to draw a distinction based on the degree of effectiveness accorded by physicians to particular CAM therapies. The issue of effectiveness occupies a central role in policy debates surrounding these practices and academic investigation into use of CAM.^26–31 Previous research has shown that patients view their physicians as important sources of information regarding the safety and effectiveness of CAM.³² Family physicians’ assessments of effectiveness provide the conceptual basis for a classification of CAM that remains focused on the issue of effectiveness while bracketing out cultural debates over the standards of effectiveness.³³

The questionnaire was designed to be as parsimonious as possible, focusing on the issues of effectiveness and general attitudes toward CAM, while collecting some sociodemographic information. Building on and updating earlier Canadian work on the subject, cited above, this research used a 10-point Likert scale and asked physicians to rate their “belief in the degree of therapeutic effectiveness” of 15 modalities of CAM:

These 15 modalities include the 14 contained in Statistics Canada’s Canadian Community Health Survey (CCHS), with the grouped practices of “Homeopathy or Naturopathy” and “Feldenkrais or Alexander Technique” remaining as such to facilitate comparison with the CCHS data. Because the conceptual rationale for treating spiritual healing as distinct from religious healing was unclear, these 2 CAM modalities were also grouped. In order to assess the degree of conceptual overlap between acupuncture and TCM, TCM (which was not included in the CCHS) was added to the questionnaire.

The representativeness of the sample of family physicians was assessed, and descriptive statistics of physicians’ awareness and ratings of the effectiveness of 15 modalities of CAM practice were analyzed. In order to explore whether or not the data supported the proposed classification of CAM practices into the categories of “accepted” and “rejected” based upon physicians’ perceptions of effectiveness, factor analysis was applied to the 15 CAM therapies family physicians were asked to rate. Factor analysis is a data reduction technique used to detect the underlying structure (dimensions) in the relationships among variables in a data set.³⁴ Subsets of variables that group together in terms of their patterns of covariation are identified via linear combinations that maximize the amount of explained variance among the variables. The components or factors thus generated are thought to be representative of the underlying structure responsible for the observed correlations among variables. In keeping with the exploratory nature of this research, the principal components method of factor analysis was used to estimate the number of components (factors) that would extract the most variance from the ratings of the 15 modalities of CAM studied.

Principal components analysis is not based on any theoretical assumptions regarding the dimensionality underlying the data structure. Factor analysis is an iterative approach in which the first factor explains the highest proportion of variance. This variance is then removed and a second linear combination that explains the maximum proportion of the remaining variance is sought, and so on.³⁴ Although there are numerous approaches that can be used to assess the dimensionality underlying a set of variables,³⁵ in order to determine the number of meaningful factors to retain, variables with eigenvalues of 1.0 or higher were identified and compared with an inspection of a scree plot for breaks among the eigenvalues.³⁶ The rotated factor pattern from principal components analysis (orthogonally rotated with varimax rotation) was used to assess the groupings of the variables into subsets and the contribution of each variable to the identified factors. Variables that loaded highly (> 0.5) and uniquely on a single factor were judged to contribute to that factor and represent the degree of correlation between that variable and the factor.

Table 2 shows that acupuncture, massage therapy, and chiropractic care were the practices with which the physicians were most familiar and to which they were most likely to assign effectiveness ratings. Less than a third of physicians knew of Feldenkrais or Alexander technique and Rolfing therapies.

As in previous Canadian studies,^19,23 this survey found that physicians believed that acupuncture, massage therapy, and chiropractic care were the most effective CAM therapies; however, they only believed this to be true of these therapies when they were used for musculoskeletal indications or, in the case of acupuncture, pain management. The mean effectiveness ratings for these 3 practices fell by about half when used for other indications. Perhaps more surprising is the mean effectiveness rating of 5.1 for spiritual or religious healing. Herbal medicine and TCM had effectiveness ratings just below the scale’s midpoint. The lowest mean effectiveness scores were for chiropractic care for nonmusculoskeletal indications, reflexology, Rolfing, Feldenkrais or Alexander technique, and homeopathy or naturopathy. The mean score for acupuncture when used for indications other than pain was the same as that for TCM (4.5). The mean effectiveness score for acupuncture used for pain management, however, was the highest reported for any CAM practice (6.7).

The factor loadings reported in Table 3 show the degree of correlation between the individual CAM practices and the extracted factors. The varimax rotated principal components analysis produced a 2-factor solution that accounts for 57.7% of the variation in the physician ratings of the effectiveness of the 15 modalities of CAM practice. (Other methods of factor analysis, such as unweighted least squares and maximum likelihood, were tried and produced similar results. Here, only the principal components results are reported.) Seven CAM practices loaded heavily and uniquely on the first extracted factor conceptualized as “rejected,” in which the most-rejected practices were chiropractic care for nonmusculoskeletal indications, homeopathy or naturopathy, massage therapy for nonmusculoskeletal indications, reflexology, herbal medicine, and finally, TCM and acupuncture for indications other than pain. This subset of variables has an eigenvalue of 3.76 and explains approximately 29% of the observed variance in the 15 CAM practices rated for effectiveness. Biofeedback, relaxation therapy, acupuncture for pain management, massage therapy for musculoskeletal indications, chiropractic care for musculoskeletal indications, and spiritual or religious healing loaded heavily and uniquely on the second extracted factor conceptualized as “accepted.” This factor also accounts for approximately 29% of the variance in the physicians’ ratings of effectiveness. With Cronbach α of .88 and .82, respectively, the therapies grouped into rejected and accepted practices demonstrated high internal consistency. The content validity of the proposed classification also makes conceptual sense in light of the qualitative data collected by this research (data not reported) and the findings of the previous research cited above, and is reflected in the mean effectiveness ratings attached to the CAM modalities by the physicians in this survey.