Dear Dr. Hamilton,
Thank you for your review of our paper on April 13th, 2017.
You highlight some important points.
Some of these points were brought up during the peer review process and these were some of our responses.
Comments regarding confounding:
If the concern is that the patient has an underlying condition such as asthma, the SCCS design incorporates a within-person comparison. So, if the patient had asthma for 5 years, this is held constant – and is completely controlled for in the analysis. The patient had asthma at the time of the adverse event and the patient had asthma during the comparison (control) period. However, if the patient just recently developed asthma immediately after receiving the corticosteroids, then asthma would not be controlled for. In that case, though, the asthma developed after giving the corticosteroids so that having concurrent asthma was not an “indication” for prescribing the drug. The SCCS design is very powerful in controlling for underlying comorbidities in a patient. If there are immediate changes in the patient’s underlying comorbidities at the time of the prescription, then there would be a concern.
Furthermore, we performed an analysis to determine the likelihood that we were detecting adverse events as a result of “being ill” rather than exposure to corticosteroids. For this analysis, we compared 30-day rates of hospitalization for sepsis, VTE, and fractures following a clinic visit in patients with matched diagnoses who did not receive corticosteroids to those who did receive corticosteroids. In this “between person” comparison, we found that rates of adverse events were consistently higher among those who received corticosteroids. It is important to note that this finding loses the advantages of the self-controlled case series design in which each patient serves as her (or his) own control. However, it provides further evidence that our findings are unlikely to be due to increased surveillance after being ill. This is shown in Table 5 of the supplemental section.
In addition, we examined the pattern of adverse events across the indications for treatment and, importantly, found consistency regardless of the indication using the SCCS (within-person) design. For this analysis, we created two groups of patients based on their indication for receiving corticosteroids: 1) respiratory conditions and 2) musculoskeletal conditions. As the reviewer suggests, sepsis might be more common after respiratory conditions if a misdiagnosis of infection occurred; however, this would not explain the higher rates of VTE or fractures after corticosteroid exposure that we also discovered. Similarly, we noted higher rates of sepsis and VTE that would not be expected after a misdiagnosis of fracture for musculoskeletal conditions like back pain. We believe the consistency of our finding for adverse events across indications supports our hypothesis of a causal association. Table 4 of the manuscript.
Regarding your concern about mechanisms of action and biological plausibility, references have been provided regarding the biological plausibility of acute effects of corticosteroids even with short durations of treatment on key pathophysiological processes. Specifically, an article by Ton et al. showed that even a very low-dose of prednisone can change indices of bone formation and bone resorption. Pouw et al. in their article in BMJ reported that even inhaled corticosteroids with much lower levels of systemic absorption reduce biological markers of bone formation. Additional references have also been provided for VTE and Infections in the original manuscript.
We agree that the balance of benefit versus risk is important when considering corticosteroids. We list the incidence rates of adverse events for corticosteroid users and nonusers, in Table 2. From these figures, the NNH can be naively calculated for the overall sample (1178 for sepsis, 461 for venous thromboembolism and 141 for fractures). However, these figures capture all events across all presenting conditions in a year. This is less relevant for individuals who are taking a short course of corticosteroids and presumably have a transient risk from this exposure. To make this more consistent for a single clinic visit within a defined time period, one could also calculate the rates in the 5 to 90 days after a visit (only in those patients with visits). When limited to this group, the comparison rates for steroid users versus nonusers for sepsis were 0.05% versus 0.02%, for venous thromboembolism were 0.14% versus 0.9%, and for fracture were 0.51% versus 0.39% with corresponding NNHs of approximately (sepsis 3333, venous thromboembolism 2000, and fracture 833). Of course, these estimates are not from the SCCS analysis which was our key method for addressing confounding and do not take into consideration the likelihood of wide individual variation in these risks.
Overall, we believe the answers we provide above raise additional concerns when translating these findings into a simple NNH for clinicians. Although we understand the value of such simplicity in messaging our findings, we are concerned that they discount much of the complexity in the relationship between corticosteroids and adverse events.
In summary, we believe there are short-term risks of taking these medications. Additional studies are needed to confirm these findings and to evaluate optimal dosing of steroids. We believe physicians should consider using alternative therapies when available.
1) Ton FN, Gunawardene SC, Lee H, Neer RM. E ects of low-dose prednisone on bone metabolism. J Bone Miner Res 2005;20:464-70. doi:10.1359/JBMR.041125.
2) Pouw EM, Prummel MF, Oosting H, Roos CM, Endert E. Beclomethasone inhalation decreases serum osteocalcin concentrations. BMJ 1991;302:627-8. doi:10.1136/bmj.302.6777.627.
Competing interests: No competing interests