To the Editor: 

I wish to thank Prinsen and colleagues for their response¹ to my recent letter to the editor,² which commented on their observational study of osteopathic manipulative treatment (OMT) in patients with low back pain.³ Although we agree on several key points, including the value of OMT in managing low back pain and the need for more pragmatic alternatives to randomized controlled trials, I am obliged to comment further on several aspects of their response. 

First, the authors stated that it would have been impossible for them to reference the OSTEOPAThic Health outcomes In Chronic low back pain (OSTEOPATHIC) Trial findings that I reported in my letter² because the results were either unpublished or in press when their manuscript was accepted for publication. However, a careful review of the chronology of the relevant publications reveals that their contention is incorrect. The primary results of the OSTEOPATHIC Trial were published in the March/April 2013 issue of Annals of Family Medicine⁴ and were freely available via the journal's website. The subgroup results for patients with moderate to severe baseline low back pain severity⁵ were initially published online in Manual Therapy, and the article was indexed as an "epub" in the National Library of Medicine's PubMed database on June 10, 2013, and was available at that time via open access. Thus, the OSTEOPATHIC Trial findings were easily accessible prior to the authors' submission of their revised manuscript on August 24, 2013. Moreover, the authors erroneously claimed that I frequently cite the results of a Consortium for Collaborative Osteopathic Research Development–Practice-Based Research Network (CONCORD-PBRN) observational study⁶ to support my comments relative to the OSTEOPATHIC Trial. 

Although Prinsen and colleagues¹ acknowledged that 45% of their pain outcome measures were missing, they provided a lengthy discussion in their response of reasons why imputation of these missing data may have been undesirable, including that Consolidated Standards of Reporting Trials (CONSORT) guidelines for reporting clinical trials discourage using the last-observation-carried-forward method of imputation.⁷ However, the updated CONSORT guideline extension for patient-reported outcomes (such as the missing pain scores in question) states that "statistical approaches for dealing with missing data should be explicitly stated for patient-reported outcomes prespecified as primary or important secondary outcomes."8 Although no method of imputation is perfect, given the large proportion of missing data it would have been prudent to perform 1 or more types of imputation in a sensitivity analysis to determine the concordance of the results with those that they observed. As it is, the validity of the results remains in question and the impact of their study is diminished by not addressing these missing data in a manner consistent with the CONSORT guideline for patient-reported outcomes.⁸ 

Prinsen and colleagues¹ stated that they addressed the potential bias attributable to clustering in their original publication; however, a word search of their article for "cluster" or "clustering" failed to detect any use of these terms. Moreover, their response contends that clustering would theoretically increase the type I error rate while simultaneously decreasing the "strength of the statistical difference," presumably between patients receiving and not receiving OMT. In reality, the type I error rate is increased because intracluster correlation of results within participating sites artificially decreases variance in the data (eg, decreases standard errors) and thereby increases the statistical significance of differences between study groups when performing hypothesis testing without adjustment for clustering. Interestingly, while disregarding the decreased variance and standard errors attributable to clustering, the authors invoked "under-estimated standard errors" as an argument against performing imputation for missing data on pain outcomes.¹ 

Finally, in clinical practice, analgesic medication is usually prescribed in response to antecedent pain and its severity. The findings reported by Prinsen and colleagues³ and by Andersson and colleagues⁹ of decreased analgesic medication use in patients receiving OMT relative to those not receiving OMT, without corresponding disparities in pain status between these 2 groups, do not appear to be plausible. Observational studies in particular, and even randomized controlled trials at times, are vulnerable to biases that may yield such implausible results. Lack of physician blinding to OMT use in the American Osteopathic Association Clinical Assessment Program could have easily biased analgesic prescribing patterns in the study by Prinsen and colleagues.³ Their inability to demonstrate the superiority of OMT over conventional treatment in decreasing pain was unlikely due to insensitivity of the visual analog scale used, as had been suggested by the authors.³ In fact, the National Institutes of Health Task Force on Research Standards for Chronic Low Back Pain now recommends a comparable numerical rating scale to measure pain.¹⁰ The best explanation for greater medication use by control patients not receiving OMT in the study by Andersson and colleagues⁹ is the excessively large number of standard medical care visits that were artificially imposed upon their physicians by the research protocol (8 visits over 12 weeks).¹¹ 

Notwithstanding many osteopathic studies performed over more than 3 decades, including that of Andersson and colleagues,⁹ the OSTEOPATHIC Trial has been recognized as a large high-quality clinical trial¹² that fills an important gap in the evidence for OMT.¹³ Unlike these previous studies, the OSTEOPATHIC Trial has shown straightforwardly that low back pain severity is significantly decreased with OMT as compared with control treatments, and that it is this improvement in pain after OMT that results in significantly decreased use of prescription analgesic medication. (doi: 10.7556/jaoa.2014.167)