To The Editor: We are writing with regard to the article “The Effect of Cyclooxygenase-2 Inhibitors on Spinal Fusion” (2002;84:1763-8), by Long et al. While we believe that the authors’ attempt to examine this important topic is laudable, their study has led them to make erroneous conclusions.

Two studies published in June 2002 clearly demonstrated the relative roles of cyclooxygenase-1 (COX-1) and cyclooxygenase-2 (COX-2) in fracture repair. In the study by Simon et al., treatment with COX-2 inhibitors was shown to lead to high rates of nonunion in a rat fracture model, whereas treatment with indomethacin did not1. Cox-2 inhibition was also associated with loss of fixation. The authors went on to show that Cox-2 (-/-) mice had deficient fracture-healing, whereas COX-1 (-/-) mice appeared to heal normally. The simultaneous publication of the study by Zhang et al. demonstrated that COX-2 (-/-) mice had grossly deficient fracture-healing and COX-1 (-/-) mice healed normally2. Furthermore, bone marrow stromal cell cultures obtained from COX-2 (-/-) mice and wild-type mice indicated that bone nodule formation was reduced 50% in COX-2 (-/-) mice. The defect in osteogenesis was completely rescued by the addition of prostaglandin E2 to the cultures. The conclusion from these studies can only be that COX-2 is essential for fracture-healing and COX-1 is not.

Both celecoxib and indomethacin block COX-1 and COX-2, but with different affinities. With use of the ratio of the COX-1 IC(50) values over the COX- 2 IC(50) values in a human whole blood assay, selectivity ratios for the inhibition of COX-2 are 6.6 for celecoxib and 0.4 for indomethacin3. In their study, Long et al., by giving ten times the clinically relevant dose of indomethacin (1 mg/kg/day), may have administered a dose sufficient to block COX-2, leading to decreased healing of the spinal fusion. The dose of celecoxib was only approximately two times the recommended dose. While celecoxib is COX-2 selective, it is possible to block COX-2 with a nonselective drug such as indomethacin if the dose is high enough.

Long et al. showed that fourteen of twenty-two spines fused after treatment with saline solution compared with ten of twenty-two that had treatment with celecoxib. This represents a 29% decrease in fusion rates with celecoxib treatment, but the power of the study to detect such a difference was only 0.23; adequate power is regarded as 0.8 (SamplePower; SPSS, Chicago, Illinois). This is a clear type-II error—that is, because of insufficient sample size, the results indicate no difference when there may well be a difference. The fusion rate in the indomethacin group dropped to four of twenty-two, a decrease of 72%; the power of the study was 90% to detect this very large difference. However, a decrease in the fusion rate of 29% would be clinically relevant to most surgeons.

Interestingly, although the study was clearly independent of pharmaceutical company involvement, the article was sent to me by a pharmaceutical company representative, who sent it, I believe, in response to information indicating that COX-2 function was essential for skeletal repair that I had passed to my orthopaedic colleagues. Long et al. unfortunately went too far in the interpretation of their limited data and stated that the effect was likely through COX-1 and not COX-2. Unless Long et al. wish to postulate that cyclooxygenase function is totally reversed in spinal fusion compared with that in fracture repair, for which they provide no evidence, it would be advisable for Long et al. and The Journal to recant this erroneous conclusion. David G. Little, MD Rick Bransford, MD Corresponding author: David G. Little, MD Westmead Children’s Hospital Orthopaedic Research and Biotechnology Department of Orthopaedics Locked Bag 4001 Westmead NSW 2145, Australia davidl3@chw.edu.au

We appreciate the letter by Drs. Little and Bransford regarding our investigation. After reviewing the articles by Simon et al as well as Zhang et al, we believe that cyclooxygenase function may indeed be quite different in a posterolateral spine fusion model compared to long bone fracture repair and that it is erroneous to equate the two. As confirmed by the Simon article, long bone fracture healing in this type of model occurs by endochondral ossification, which these publications suggest may be dependent upon COX-2. Lumbar intertransverse process fusion after iliac crest bone grafting in the rabbit occurs mainly through membranous bone formation. We do not claim that our results are extrapolatable to a fracture model which is known to repair itself through endochondral ossification. Even within one species, the results of a study on bone formation and healing in one anatomical location are not necessarily generalizable to another. For example, the effect of BMP in lumbar interbody fusions can not be used to make reasonable conclusions about posterolateral lumbar fusions in even the same patient. Even in the same species, with the same type of ossification, there are puzzling differences. Simon, et al, found that in COX-2 -/- mice, fetal and developmental endochondral ossification occurs normally, while long bone fracture endochondral ossification is adversely affected. This suggests that even the role of COX-2 in the endochondral ossification of the same animal at different stages of life or under different conditions belies a simple explanation. The anatomical area, the use of bone graft, the particular animal model used, the interval between the experiment and sacrifice, and the doses of drugs used may all affect the final results of any study when one is investigating bone healing. We feel that because all of these factors differed in the two studies as compared to ours, the results of these studies are neither comparable nor contradictory. We agree that the lack of statistical significance in our palpation results may have been due to a lack of power. This is why we stated in our discussion that “there was a trend toward significance for the difference between both the gross and histologic findings. It may be that if we had studied more animals there would have been significant differences.” We found that 14 of 22 control spines (64%) and 10 of 22 Celecoxib spines (45%) were fused upon gross inspection and palpation. This represents a difference of 19%, not 29%. However, for the radiographic assessment, 18 of 22 (82%) controls, 19 of 22 (86%) celecoxib but only 9 of 22 (41%) indomethacin rabbits were felt to be fused. With the histological analysis, the average scores were 5.2 for the controls, 4.8 for the celecoxib group and 3.5 for the indomethacin group. The sum of all three methods suggests a greater difference between indomethacin and control than celecoxib and control.

In the long bone fracture model, the authors found that at a dose of indomethacin one tenth of ours, indomethacin does not prevent long bone healing. We respectfully disagree with this reduced dose as being the clinically relevant dose. But let us assume that these authors indeed used the “normal” dose and we overdosed by a factor of 10. Our dose would have blocked both COX-1 and 2, and either may have resulted in the lowered fusion rates seen with indomethacin. But if COX-2 blockade were responsible, the COX-2 inhibited group (celecoxib) should have shown a lower fusion rate also. Celecoxib is 6.6 times more likely to block Cox-2 over Cox-1. Drs. Little and Bransford state that we gave our animals a dose comparable to two times the recommended human dose. At this dose, all of Cox-2 should have been blocked, while only some of COX-1 should have been blocked. If both groups had complete COX-2 blockade, then the only difference between the two was that the celecoxib group had only a partial COX-1 blockade, while the indomethacin had complete COX-1 blockade. We stand by our belief that the most logical interpretation of this data is that the difference in the fusion rates is due to the difference in the degree of COX-1 inhibition. Finally, we would like to emphasize that we do not advocate the use of any anti-inflammatories following lumbar intertransverse process arthrodeses in humans. We strongly believe that all anti-inflammatories should be avoided if at all possible following spinal arthrodesis until a solid fusion is obtained. If an anti-inflammatory must be used, our results in rabbits suggest that a COX-2 specific one may be preferable to indomethacin.

K. Daniel Riew, M.D. Associate Professor Chief, Cervical Spine Surgery Department of Orthopaedic Surgery Barnes-Jewish Hospital & Washington University School of Medicine

We write to reply to the recent letter by Dr.Benjamin A. Goldberg

During the process of designing our study we realized there were no published dosages of Celecoxib in rabbits. In lieu of having this information, we performed an allometric scaling calculation to estimate a rabbit dose (10 mg/kg/day) that would be consistent with the recommended anti-inflammatory dose for humans. We realize that performing this study, in the absence of a clear pharmacokinetic analysis, was a potential weakness of the study. However, the dose used was based upon scientifically valid calculations. We are not reporting that COX-2 inhibitors absolutely have no effect on bone healing. Rather, we are reporting that Celecoxib at a dose of 10 mg/kg/day in the rabbit spine fusion model did not show a significant difference when compared to vehicle control. In future studies utilizing the rabbit model, it would be prudent to elucidate the pharmacokinetic data. Until that information is available, we firmly stand behind our study design and conclusions.

Most of the studies evaluating the effect of COX-2 inhibitors were using rodent models. Dr. Goldberg, in his critique, refers to studies evaluating COX-2 inhibitors in rat models. Much of the current research in rat models utilize well accepted anti-inflammatory dosages of COX-2 inhibitors. However, attempts to extrapolate data from rat models and apply to rabbit models may be misleading due to potential species differences in bone healing and drug metabolism.

To the Editor:

We read with interest “The Effect of Cyclooxygenase-2 Inhibitors on Spinal Fusion” (2002, 84:1763-1768) by Long, Lewis, Kulko, Zhu, and Riew. The authors concluded that celecoxib does not significantly inhibit the rate of spinal fusion in rabbits. In addition, the authors attributed the adverse effects of non-selective non-steroidal anti-inflammatory drugs (NSAIDs) to the inhibition of the COX-1 enzyme. Their results and conclusions are quite contradictory to other research on the effects of COX-2 inhibitors on bone healing.1,2,3

Our research reported that rats with experimentally created non- displaced femoral fractures taking rofecoxib were significantly more likely to have clinical non-unions (p <_0.0001 reduced="reduced" radiographic="radiographic" healing="healing" maturity="maturity" p="p" increased="increased" average="average" fracture="fracture" angulation="angulation" greater="greater" mean="mean" callus="callus" width="width" and="and" a="a" lower="lower" histological="histological" grade="grade" at="at" four="four" weeks="weeks" post-fracture="post-fracture" compared="compared" to="to" control="control" rats.="rats." further="further" rats="rats" taking="taking" ibuprofen="ibuprofen" had="had" similar="similar" adverse="adverse" effects="effects" on="on" bone="bone" consistent="consistent" with="with" previously="previously" published="published" studies.45="studies.45" goodman="goodman" et.="et." al.="al." simon="simon" noted="noted" detrimental="detrimental" of="of" cox="cox" _-2="_-2" inhibitors="inhibitors" formation.="formation." /> An explanation for these contradictory results and conclusions may be that the rabbits in the study by Long et. al. did not receive a therapeutic dose of celecoxib. The authors noted that there is not a published therapeutic dose for rabbits and thus used allometric scaling calculations6 to calculate the therapeutic dose of celecoxib as 10 mg/kg/day. Allometric scaling as described by Timm et. al. was used to extrapolate surface area to volume relationships of different size but similarly shaped (cylindrical), cold-blooded reptiles of the same species (snakes). It is unknown whether these calculations can be applied to non- cylindrically dissimilarly shaped species (human and rabbit) of different body temperatures and potentially variable metabolism of pharmaceutical agents. The package insert for Celebrex7states that the two-fold human exposure in male rats is 200 mg/kg/day based on AUC0-24 of serum concentrations in humans and rats. This would suggest that the 10 mg/kg/day for rabbits as selected by the authors may be too low to expect an observable effect on spine fusion since there were was no serum concentration measurements celecoxib. Had the selected dosage been higher, the authors may well have found a delay in healing comparable to the ibuprofen group.

Our study used adult male rats because of the extensive availability of pharmacokinetic data on COX-2 inhibitors in rodents.7,8 Rats underwent intramedullary fixation of a non-displaced closed femoral fracture as described by Bonnarens and Einhorn9. They were given a dose of 8 mg/kg/day of rofecoxib to simulate two-fold human exposure based on AUC0- 24 of serum concentrations in humans and rats. At that dosage level, we found that rofecoxib produced a significant effect in delaying bone healing.

As COX-2 is induced in injury and inflammation10,11, our results and those of Goodman et. al. and Simon et. al. suggest that COX-2 inhibitors should be used with caution in situations that require bone healing.

1 Goodman SB, Ma T, Ikenoue T, Matsura I, Trindade M, Fox N, Wang N, Genovese M, Smith RL: COX-2-Selective NSAID Decreases Bone Ingrowth in Vivo. 48th Annual Meeting of the Orthopaedic Research Society, Transaction #0066, Vol. 27, Dallas, TX, February 2002.

2 Simon AM, Sabatino CT, O’Connor JP: Effects of Cyclooxygenase-2 Inhibitors on Fracture Healing. 47th Annual Meeting of the Orthopaedic Research Society, Transaction #0205, Vol. 26, San Francisco, CA, 2001.

3 Leonelli S, Goldberg B, Safanda, J, Bagwe M, Sethuratnam, S King, S: The Effect of Cyclooxygenase-2 (COX-2) Inhibitors on Bone Healing. 48th Annual Meeting of the Orthopaedic Research Society, Transaction #0065, Vol. 27, Dallas, TX, February 2002.

4 Huo MH, Troiano NW, Pelker RR, Gundberg CM, Friedlaender GE: The influence of ibuprofen on fracture repair: biomechanical, biochemical, histological, and histomorphometric parameters in rats. J Orthop Res 1991; 9(3): 383-90.

5 Altman RD, Latta LL, Keer R, Renfree K, Hornicek FJ, Banovac K: Effect of nonsteroidal antiinflammatory drugs on fracture healing: a laboratory study in rats. J Orthop Trauma 1995; 9(5): 392-400.

6 Timm KI, Picton JS, Tylman B: Surface area to volume relationships of snakes support the use of allometric scaling for calculating dosages of pharmaceuticals. Lab Anim Sci, 1994; 44: 60-2

7 G.D. Searle & Co., Celebrex Package Insert. 1998.

8 MERCK & Co., Vioxx Package Insert. 1998.

9 Bonnarens F, Einhorn TA: Production of a standard closed fracture in laboratory animal bone. J Orthop Res, 2: 97-101, 1984.

10 Urban MK: COX-2 specific inhibitors offer improved advantages over traditional NSAIDs. Orthopaedics (Thorofare, NJ), 23: S761-4, 2000.

11 Smith CJ, et. al. : Pharmacological analysis of cyclooxygenase-1 in inflammation. Proc Natl Acad Sci USA, 95: 13313-8. 1998.