Effects of doxycycline on progression of osteoarthritis: Results of a randomized, placebo-controlled, double-blind trial




To confirm preclinical data suggesting that doxycycline can slow the progression of osteoarthritis (OA). The primary outcome measure was joint space narrowing (JSN) in the medial tibiofemoral compartment.


In this placebo-controlled trial, obese women (n = 431) ages 45–64 years with unilateral radiographic knee OA were randomly assigned to receive 30 months of treatment with 100 mg doxycycline or placebo twice a day. Tibiofemoral JSN was measured manually in fluoroscopically standardized radiographic examinations performed at baseline, 16 months, and 30 months. Severity of joint pain was recorded at 6-month intervals.


Seventy-one percent of all randomized subjects completed the trial. Radiographs were obtained from 85% of all randomized subjects at 30 months. Adherence to the dosing regimen was 91.8% among subjects who completed the study per protocol. After 16 months of treatment, the mean ± SD loss of joint space width in the index knee in the doxycycline group was 40% less than that in the placebo group (0.15 ± 0.42 mm versus 0.24 ± 0.54 mm); after 30 months, it was 33% less (0.30 ± 0.60 mm versus 0.45 ± 0.70 mm). Doxycycline did not reduce the mean severity of joint pain, although pain scores in both treatment groups were low at baseline and remained low throughout the trial, suggesting the presence of a floor effect. However, the frequency of followup visits at which the subject reported a ≥20% increase in pain in the index knee, relative to the previous visit, was reduced among those receiving doxycycline. In contrast, doxycycline did not have an effect on either JSN or pain in the contralateral knee. In both treatment groups, subjects who reported a ≥20% increase in knee pain at the majority of their followup visits had more rapid JSN than those whose pain did not increase.


Doxycycline slowed the rate of JSN in knees with established OA. Its lack of effect on JSN in the contralateral knee suggests that pathogenetic mechanisms in that joint were different from those in the index knee.

In the present report, we describe the results of a randomized, placebo-controlled, double-blind trial of the tetracycline antibiotic, doxycycline, in subjects with knee osteoarthritis (OA). Selection of doxycycline as a potential disease-modifying OA drug was based not on the premise that OA is an infectious disease, but rather on results of in vitro studies showing 1) that doxycycline inhibited the degradation of type XI collagen, one of the minor collagens of articular cartilage, by 72-kd gelatinase (1); 2) that the presence of doxycycline during activation of procollagenase resulted in generation of low molecular weight, catalytically inactive fragments and marked reduction in the levels of active enzyme (2); and 3) that doxycycline inhibited messenger RNA for inducible nitric oxide synthase, an enzyme present in large quantities in OA cartilage, the activity of which results in secretion of matrix metalloproteinases (MMPs) by the chondrocyte (3, 4).

With this background, combined with evidence from Golub et al (5) that oral administration of tetracycline prevented periodontal disease in diabetic rats via inhibition of gingival collagenase, investigators in our group undertook in vivo studies of the effect of doxycycline in a canine cruciate-deficiency model of OA. These studies showed that this drug significantly reduced levels of active and total gelatinase and collagenase in extracts of the OA cartilage (6). Similar reductions in cartilage collagenase and gelatinase were obtained after administration of doxycycline to humans undergoing total joint arthroplasty (7). Furthermore, doxycycline administration markedly reduced the incidence and progression of joint pathology in the above-mentioned canine model of OA (6), which was consistent with evidence that it reduced the severity of spontaneous OA in the Hartley-Dunkin strain of guinea pig (8) and that treatment with a chemically modified tetracycline reduced the severity of joint pathology in a surgically induced model of OA in rabbits (9).

The primary outcome measure in this trial was the rate of joint space narrowing (JSN) in the medial tibiofemoral compartment. Based in part on epidemiologic evidence from the Chingford Health Study (10), a purportedly high-risk target population was selected (i.e., obese, middle-aged women with unilateral knee OA viewed on the conventional standing anteroposterior [AP] radiograph) that permitted co–primary comparisons of the treatment groups with respect to JSN in the index knee (which exhibited radiographic evidence of established OA at baseline) and in the contralateral knee (in which radiographic evidence of tibiofemoral OA in the standing AP view was absent at baseline). Secondary outcome measures included knee pain and function.


The procedures, benefits, risks, and safeguards in this trial were approved by the Institutional Review Boards affiliated with the 6 participating Clinical Research Centers (Indiana University—Purdue University at Indianapolis, Northwestern University, University of Alabama at Birmingham, Arthritis Research Center Foundation, University of Arizona, and University of Pittsburgh). A data and safety monitoring board convened semiannually throughout the study.


All subjects were obese women ages 45–64 years with unilateral radiographic knee OA according to the criteria of the American College of Rheumatology (11), with Kellgren/Lawrence (K/L) grade 2 or 3 changes in one knee (the index knee) and grade 0 or 1 changes in the contralateral knee (12) in a conventional standing AP view. All subjects were in the upper tertile of the age- and race-adjusted norms for body mass index (BMI) in women (13).

Exclusion criteria included posttraumatic or any other form of secondary knee OA, the presence of inflammatory arthritis, comorbidity that would confound measurement of knee pain and/or function, menopausal/contraceptive status that did not preclude pregnancy, and a history of tetracycline allergy. Subjects were also ineligible for enrollment if they had received an intraarticular injection of hyaluronan within the previous 6 months or of corticosteroid within the previous 3 months.

Run-in period.

Prior to randomization, all eligible subjects received a single-blind, 30-day supply of placebo capsules and underwent a 4-week, single-blind run-in test (14, 15) (Figure 1), which was designed to exclude from randomization subjects who were incapable of adhering (or unwilling to adhere) to a twice-daily dosing regimen and/or who could not/would not keep scheduled appointments. To “pass” the run-in test, subjects were required to keep 2 biweekly appointments (at week −2 and week 0) and to maintain ≥80% adherence to a twice-daily dosing regimen over the entire 4-week period (as indicated by the dates on/times at which each dose was taken) as recorded in the memory of an electronic dosing monitor (Medication Event Monitoring System; AARDEX, Union City, CA) contained in the cap of the medication vial (16, 17). The requirements for the run-in test (and the consequences of failing) were disclosed to the subject prior to screening, at the time informed consent was obtained.

Figure 1.

Summary of accrual of subjects, discontinuation of study drug, and subject retention for the analysis of 30-month radiographic outcomes. Random assignment of subjects to treatment groups was preceded by a 4-week run-in period, during which each otherwise qualified subject was required to keep appointments at week −2 and week 0 and to maintain ≥80% compliance (therapeutic coverage) with a twice-daily dosing regimen (of placebo). Frequencies of reasons for discontinuation of treatment in the double-blind phase of the trial reflect the judgment of the investigator regarding the relationship of the adverse event (AE) to doxycycline (Doxy) before the blind was broken.

Double-blind phase.

Subjects who satisfied the requirements for the run-in test were randomly assigned to receive 100 mg doxycycline or matched placebo twice a day. Within each clinical center, subjects were allocated randomly to treatment groups in blocks of 6. Subjects were permitted to continue taking any prescription or over-the-counter pain medication for their knee OA throughout the trial, except during scheduled washout periods (see below). The double-blind phase entailed a baseline visit and 15 bimonthly followup visits, during each of which adherence to the treatment regimen (therapeutic coverage) and the occurrence of adverse events were recorded. Adherence was quantified cumulatively, with the subject receiving credit for 16 hours of coverage following each dose of study drug recorded by the electronic monitor. At the screening visit and every 6 months after randomization, blood and urine samples were obtained for the performance of laboratory tests for safety, which included measurement of serum levels of alkaline phosphatase, aspartate aminotransferase, alanine aminotransferase, bilirubin, and creatinine, as well as a complete blood cell count with differential and a urinalysis.

Radiographic studies.

In addition to the conventional standing AP view that was used to determine eligibility, a fluoroscopically standardized semiflexed AP view (18), a supine lateral view of each knee, and skyline (Hughston) (19) views of the patellofemoral joints were obtained at baseline and 16 months and 30 months later. A conventional standing AP radiograph was also obtained at the final study visit.

The primary outcome measure was JSN in the medial tibiofemoral compartment, a surrogate for thinning of articular cartilage. The minimum joint space width (JSW) in the medial compartment was measured manually in the semiflexed AP view, according to the method of Lequesne (20), using the points of a screw-adjustable compass and a graduated magnifying lens. Measurements were made by an observer (SAM) who was blinded to the treatment group assignment of the subject. Measurements of JSW were adjusted for radiographic magnification, based on measurement of the image projected onto the film by a metallic ball (6.35 mm in diameter) that was affixed to the skin over the head of the fibula during the examination. The intra- and interreader reproducibilities of repeated measurements of JSW in a random sample of 30 radiographs (on which all identifying information was masked) were excellent (intraclass correlation coefficients of 0.99 and 0.96, respectively) (21).

Although the procedure for measurement of JSW was not blinded to the sequence of the radiographs (i.e., baseline, month 16, month 30), given that 1) each estimate of magnification-corrected JSW was a compound measurement of minimum interbone distance and the diameter of the magnification marker, 2) the Lequesne method (20) separates the recording of distances to be measured (which are obtained as pinpricks on a blank sheet of paper) from the actual measurement of those distances, and 3) the reader was blinded to the treatment group, it is highly unlikely that the absence of blinding to sequence permitted any bias regarding the comparison of active treatment and placebo groups with respect to JSN.

Clinical assessments.

The clinical severity of knee OA was assessed every 6 months after a washout (5 half-lives) of all nonsteroidal antiinflammatory drugs (NSAIDs) and analgesics. Subjects were permitted to take acetaminophen, as needed, up to 4 gm/day during all but the final 24 hours of the washout period. The clinical assessment included administration of the Likert version of the pain and physical functioning scales of the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) (22). Knee pain after a 50-foot walk was measured on a 10-cm visual analog scale (VAS) (0 = no pain; 10 = extreme pain). The subject and the investigator rated global disease activity on a 5-point Likert scale (1 = none; 2 = mild; 3 = moderate; 4 = severe; 5 = extreme). With the exception of the WOMAC physical functioning scale, all elements of the clinical assessment were recorded separately for the index and contralateral knees.

Statistical analysis.

To assess the effectiveness of randomization, baseline demographic, clinical, and radiographic variables were compared between groups using t-tests for continuous measures, Wilcoxon rank sum tests for ordinal variables, and chi-square tests for nominal characteristics. The primary analysis for this study was the comparison of JSN in the doxycycline and placebo treatment groups. Group comparisons were performed using all subjects who underwent followup assessment, regardless of whether they had discontinued the study drug prematurely or completed the trial per protocol.

The null hypothesis of no effect was tested separately in the index and contralateral knees using a mixed-effects linear model (i.e., repeated-measures) approach with the critical value for statistical significance set at 0.05. The model included a random subject effect and fixed effects for treatment group, clinical center, and visit (i.e., a class variable with two levels, 16 months or 30 months), the clinical center–treatment group and visit–treatment group interactions, and baseline JSW. The model for each knee also included covariates representing demographic, clinical, or radiographic variables significantly related to JSN (e.g., pain at baseline). Comparisons between treatment groups at 16 months and 30 months were not adjusted for multiple tests because they were performed to describe the time course of the treatment effect rather than to test discrete hypotheses. To account for any possible effect of variation in timing of the 16- and 30-month visits, parallel mixed-effect linear model analyses were also performed with followup time as a continuous measure.

A mixed-effects linear model approach was also used to compare treatment groups with respect to secondary outcomes (i.e., mean scores for knee pain, function, global assessments), which were measured every 4–6 months throughout the double-blind phase. For the global assessments, which are ordinal measures, a rank transformation within visit (but across treatment) was implemented prior to fitting the mixed model.

Because a threshold level of knee pain was not among the inclusion criteria for this study, there was no assurance that pain scores would be sufficiently high to permit detection of a significant effect of doxycycline on mean pain scores. Therefore, a priori comparisons of treatment groups with respect to mean pain scores were supplemented with post hoc group comparisons of the frequency with which subjects reported clinically significant increases in knee pain across successive semiannual pain assessments. For each subject, we tallied the number of pain assessments during which knee pain was increased ≥20% (with a minimum increase of 1 cm on the VAS) compared with that reported 6 months earlier. Student's t-test was used to compare treatment groups with respect to the mean frequency (i.e., percentage of up to 5 followup pain assessments) with which ≥20% increases in knee pain were reported.

Changes in concomitant medications for OA pain, including the start and stop dates and daily doses of all prescription and over-the-counter NSAIDs and analgesic drugs (acetaminophen and opioid analgesics), were documented at each visit. Treatment groups were compared over time with respect to the types of pain medications taken for OA pain (by chi-square test) and the strength of NSAID dose, expressed as the percentage of the minimum daily dose recommended for OA pain in the Physicians' Desk Reference (23). Repeated-measures analysis of variance (ANOVA) was used to compare those who took NSAIDs in the 2 treatment groups with respect to the strength of dose.

The frequencies of adverse events in the 2 groups were compared using Fisher's exact test. Even though relatively few subjects were lost to followup, we evaluated the possible effects of missing data on the results. First, to determine whether missing data could be considered to be missing completely at random, demographic and baseline clinical characteristics of subjects who underwent their 30-month radiographic examination were compared with those of subjects who were lost to followup at 30 months, using t-tests and chi-square tests. Second, JSN data for the index knee were further examined for evidence of nonignorable dropout, using an informative censoring modeling approach (24, 25) that assumes that serial measurements of JSW follow a linear regression, with a random slope and an intercept for each subject. This method yields a test of the assumption of nonignorable dropout, in which a significant result indicates the missing data cannot be ignored. If the hypothesis is not rejected, the missing data can be considered to be missing at random and the mixed-model approach outlined above is appropriate.


Between May 1997 and May 2000, 1,975 volunteers underwent clinical and radiographic screening (Figure 1). Clinical eligibility and radiographic evidence of unilateral knee OA on the standing AP radiograph were confirmed in 489 of these volunteers (25%), 463 of whom agreed to take the run-in test. Of these, 431 subjects (93%) passed the run-in test and were randomly assigned to receive 100 mg doxycycline (n = 218) or placebo (n = 213) twice a day.

The 2 treatment groups were equivalent at baseline with respect to all demographic variables, BMI, K/L grade of radiographic severity, minimum JSW in the medial compartment, knee pain and function, and the types of drugs taken for OA pain (Table 1). Among the subjects who were randomized, 307 (71%) completed the 30-month trial per protocol (149 still taking doxycycline, 158 still taking placebo). Among the 124 who discontinued the study drug prematurely, 47 (38%) did so because of the burden of participation, 35 (28%) because of a possible adverse effect of doxycycline, 25 (20%) because of an adverse effect unrelated to doxycycline, and 17 (14%) because they moved away (Figure 1). Compared with subjects who completed the trial per protocol, dropouts had significantly higher mean baseline scores for WOMAC pain (both knees) and physical functioning (P < 0.05 in each case).

Table 1. Characteristics of subjects at baseline*
 Treatment group
Combined (n = 431)Doxycycline (n = 218)Placebo (n = 213)
  • *

    Except where indicated otherwise, values are the mean ± SD. BMI = body mass index; OA = osteoarthritis; K/L = Kellgren/Lawrence; JSW = joint space width; WOMAC = Western Ontario and McMaster Universities Osteoarthritis Index; VAS = visual analog scale; NSAID = nonsteroidal antiinflammatory drug; COX-2 = cyclooxygenase 2.

  • Graded in the standing anteroposterior (AP) view.

  • Measured in the semiflexed AP view.

  • §

    Five-point scale (1 = none; 2 = mild; 3 = moderate; 4 = severe; 5 = extreme).

  • Percentages total >100% because some subjects took >1 pain medication.

Age, years54.9 ± 5.654.8 ± 5.555.0 ± 5.8
BMI, kg/m236.7 ± 6.236.8 ± 6.336.5 ± 6.0
White, %818279
Overall severity of tibiofemoral OA, index knee, %   
 K/L grade 2596256
 K/L grade 3413844
Overall severity of tibiofemoral OA, contralateral knee, %   
 K/L grade 0686868
 K/L grade 1323232
Minimum medial JSW, mm   
 Index knee3.62 ± 1.173.64 ± 1.153.60 ± 1.19
 Contralateral knee3.94 ± 0.883.91 ± 0.903.96 ± 0.86
Index knee pain   
 WOMAC pain score, 5–2510.8 ± 4.210.8 ± 4.010.9 ± 4.3
 50-foot walk pain, 0–10-cm VAS2.1 ± 2.42.1 ± 2.42.0 ± 2.5
Contralateral knee pain   
 WOMAC pain score, 5–259.0 ± 4.18.8 ± 4.09.2 ± 4.1
 50-foot walk pain, 0–10-cm VAS1.4 ± 2.21.4 ± 2.11.5 ± 2.3
WOMAC function score, 17–8538.3 ± 13.038.5 ± 12.138.2 ± 13.8
Global disease activity, index knee, median (range)§   
 Patient's rating2 (1–5)2 (1–5)2 (1–5)
 Physician's rating2 (1–5)2 (1–4)2 (1–5)
Drugs used for OA pain, subjects taking, %   
 Nonselective NSAID606356
 Opioid analgesic786
 COX-2 selective inhibitor121

Sixty of the 124 treatment dropouts (48%) returned for a knee radiograph at 30 months, resulting in an overall rate of loss to followup of only 14.8% and enabling an analysis of 30-month outcomes in 85% of all randomized subjects. Despite the difference noted above between dropouts and completers with respect to pain and function at baseline, informative censoring models for the index knee showed that the data could be treated as missing at random (P = 0.82). The data were therefore analyzed with mixed-effect linear models. This method uses all available data and yields unbiased results similar to those obtained with methods involving the imputation of missing data (24, 25). The mean therapeutic coverage was very high in both treatment groups: 91.8% among subjects who completed the study per protocol and 81.5% among those for whom 30-month radiographic data were available (including the interval during which dropouts were not taking the study drug).

At the end of the study, radiographs of the index knees of 2 subjects in the placebo group indicated the presence of Paget's disease of bone (in 1 subject) and osteonecrosis (in 1 subject) that was not apparent in the baseline examination. Data from these subjects were excluded from the efficacy analyses. Similarly, efficacy data obtained after surgery from 4 subjects (2 in each group) who underwent total knee arthroplasty during the study were not included in the outcomes analyses, although they were included in the safety evaluation.

Quantitative measurement of JSN.

Index knee.

Preliminary analyses identified index knee pain at baseline as a significant covariate of JSN in that joint (P < 0.001 for WOMAC pain; P < 0.05 for 50-foot walk pain). Because these two pain scales were highly correlated, the former was used as a covariate in the initial repeated-measures model. The interaction terms (treatment group–clinical center, treatment group–visit) were not statistically significant in the mixed model. The 2 treatment groups differed significantly from each other with respect to overall rate of JSN (P = 0.009), visit (P < 0.001), clinical center (P = 0.002), baseline JSW (P < 0.001), and baseline pain (P < 0.001). Deviations in the timing of radiographic examinations around the 16-month and 30-month due dates were not significant; the results of the above analysis were unchanged when the discrete time variable (16 months or 30 months) was replaced with a continuous variable representing the actual interval, to the day, between baseline and followup radiographs. After 16 months of treatment, the mean ± SD loss of JSW in the index knee in the doxycycline group was 40% less than that in the placebo group (0.15 ± 0.42 mm versus 0.24 ± 0.54 mm; adjusted P = 0.027) (Table 2); after 30 months, it was 33% less (0.30 ± 0.60 mm versus 0.45 ± 0.70 mm; adjusted P = 0.017).

Table 2. Radiographic outcomes shown as quantitative changes in JSW*
 Index kneeContralateral knee
nMean ± SDnMean ± SDnMean ± SDnMean ± SD
  • *

    See Table 1 for definitions.

  • K/L grade 2 or 3 at baseline.

  • K/L grade 0 or 1 at baseline.

  • §

    P values for group comparisons with respect to changes in JSW come from repeated-measures models with treatment group, location, baseline covariates, visit, treatment group–location interaction, and treatment group–visit interaction as the independent variables. Baseline JSW and knee pain were covariates for analysis of changes in JSW in the index knee; baseline BMI was the covariate for analysis of changes in JSW in the contralateral knee.

Baseline JSW, mm2183.64 ± 1.152113.61 ± 1.190.7402183.91 ± 0.902113.96 ± 0.860.554
16-month change in JSW, mm1880.15 ± 0.421910.24 ± 0.540.0271880.23 ± 0.451910.23 ± 0.590.504
30-month change in JSW, mm1810.30 ± 0.601800.45 ± 0.700.0171810.47 ± 0.701800.41 ± 0.670.317

Contralateral knee.

Preliminary analyses of JSN in the contralateral knee identified only BMI as a significant covariate (P = 0.025) in the repeated-measures model analysis. The final mixed model identified visit and clinical center, but not treatment group, as significant sources of variation (P < 0.001 and P = 0.015, respectively). The results were unchanged when followup time was included as a continuous variable.

At both 16 months and 30 months, the mean rate of loss of JSW in the contralateral knee in the placebo group was comparable in magnitude and variability with that in the index knee (Table 2). When the subjects were divided into tertiles on the basis of baseline JSW in the contralateral knee, it was apparent that even among those in whom JSW at baseline was smallest (and whose contralateral knees therefore most closely resembled their index knees in this respect), doxycycline did not affect the rate of JSN (data not shown).

Secondary outcomes.

From a clinical perspective, mean baseline scores for the secondary outcome measures indicated that the OA in these subjects was mild (Table 1), with little room for improvement in the double-blind phase. This is not surprising, since the subjects were recruited chiefly from the general population rather than from a clinic they were attending because of knee pain, and the inclusion criteria did not require that they exceed a threshold level of joint pain.

Three hundred eighty-three subjects underwent at least 1 followup clinical evaluation (WOMAC, VAS for 50-foot walk pain, and global assessments). For both knees, mean scores for WOMAC pain, 50-foot walk pain, and WOMAC function and median scores for the patient's and physician's global assessments were similar in the 2 treatment groups at baseline and throughout the double-blind phase. Nonetheless, the treatment groups differed in the frequency with which clinically important increases in pain were observed across successive semiannual assessments. The mean percentage of pain assessments in which subjects in the placebo group reported a ≥20% increase in knee pain (WOMAC pain score), compared with that reported 6 months earlier, was 27% greater than that in the doxycycline group (mean ± SD 30.2 ± 21.1% versus 23.8 ± 22.7%; P = 0.004). Comparable treatment group differences were found in 50-foot walk pain, for which a change of at least 1 cm on the 10-cm VAS was required in order to be considered clinically significant (24.3 ± 23.2% in the placebo group versus 19.4 ± 21.7% in the doxycycline group; P = 0.032). In contrast, treatment did not have a significant effect on the frequency of increases in WOMAC pain or 50-foot walk pain in the contralateral knee.

Relationship of JSN to frequency of increases in knee pain.

In both knees, the percentage of semiannual assessments in which 50-foot walk pain was ≥20% more severe than that reported 6 months previously was directly related to the rate of JSN over 30 months (Figure 2). One-way ANOVA and post hoc analyses showed that the rate of JSN in both knees was significantly more rapid in patients reporting ≥20% increases in 50-foot walk pain during the majority (60–100%) of their followup visits than in patients who reported no increase in pain (P < 0.05 for each knee). Similar patterns were seen for WOMAC pain scores for the contralateral knee (data not shown).

Figure 2.

Association between the mean rate of joint space narrowing (JSN) and the frequency of ≥20% increases in 50-foot walk pain in successive semiannual assessments. Among subjects in both treatment groups who reported an increase in 50-foot walk pain during 3 or more (≥60%) of the 5 pain assessments, the mean rate of JSN was more than twice as rapid as that in subjects who reported no increases in knee pain.

Effects of intraarticular injections and nutritional supplements.

As noted above, subjects were permitted to follow the recommendations of their physicians regarding treatment of OA during the trial, including intraarticular injection of corticosteroids or hyaluronic acid (HA), both of which have been suggested to affect the rate of cartilage loss in knee OA (26–28). Subjects were also free to take nutritional supplements that have been associated with slower rates of progression of knee OA in epidemiologic studies (i.e., vitamin C, vitamin D) (29, 30) or clinical trials (i.e., glucosamine) (31, 32). Notably, neither injectable therapeutic agents nor nutritional supplements confounded the effects of doxycycline on JSN. Only 5 subjects (1 in the doxycycline group, 4 in the placebo group) received a corticosteroid injection during the trial; 1 of the 4 subjects in the placebo group also received a series of HA injections.

The 2 treatment groups were nearly identical with respect to the percentage of subjects who took vitamin C (54% in each group), vitamin D (49% in the placebo group and 54% in the doxycycline group), or glucosamine (24% in each group) at any time during the trial. Notably, JSN in the index knee of subjects who took glucosamine for any duration during the trial (with or without chondroitin sulfate) was more rapid than that in subjects who took no glucosamine (0.21 mm/year versus 0.13 mm/year; P < 0.05 by Student's t-test). The rate of JSN among subjects who took only glucosamine (0.24 mm/year) was similar to that in subjects who took glucosamine and chondroitin sulfate (0.20 mm/year). These observations must be viewed with caution, however, because exposure to glucosamine was uncontrolled, and glucosamine use may have been related to the severity of baseline knee pain, which was shown to be a significant baseline covariate for the rate of JSN in the index knee (see above).

Effects of analgesics/NSAIDs taken for OA pain.

As shown in Table 1, 26% of subjects took no pain medication at baseline despite the presence of knee OA. Nonselective NSAIDs were the most commonly used pain medications (60% of all subjects). The small number of individuals taking cyclooxygenase 2 (COX-2) selective inhibitors at baseline (6 subjects [1%]) reflects the fact that most subjects were enrolled prior to the availability of COX-2 selective inhibitors at the end of 1999. By the end of the study, only 10 subjects (3% of completers) were not taking pain medication. Increases compared with baseline were observed in the frequency of use of all classes of analgesics (acetaminophen, opioid analgesics, NSAIDs)—most notably in the frequency of use of COX-2 selective inhibitors (23% of completers). The dosing of NSAIDs remained fairly constant over the course of the trial (on average, ∼113–118% of the recommended minimum daily dose for OA pain). However, the 2 treatment groups did not differ with respect to the types of drugs taken for OA pain, the frequency of dosing, or the daily NSAID dose. Moreover, after controlling for baseline JSW, baseline pain, and treatment group, the frequency and dose of NSAIDs were unrelated to the rate of JSN in the index knee.

Adverse events.

The frequencies of specific adverse events reported by ≥5% of subjects during the double-blind phase, regardless of severity, are shown in Table 3. Adverse events that occurred significantly more frequently (P < 0.05) in the doxycycline group than in the placebo group were restricted to recognized side effects of doxycycline (i.e., monilial vaginitis, sun sensitivity, nonspecific gastrointestinal [GI] symptoms). However, only a small proportion of subjects reporting doxycycline-related side effects discontinued the study medication prematurely (Table 3). Subjects in the active treatment group reported fewer urinary tract infections (P < 0.05), and there was a trend toward fewer upper respiratory tract infections in the doxycycline group than in the placebo group (P = 0.09).

Table 3. Adverse events reported by ≥5% of randomized subjects*
  • *

    Values are the number (%) of patients. Concurrent nonspecific gastrointestinal side effects (e.g., nausea and vomiting) were often cited as the reason for discontinuation of study medications. ALT = alanine aminotransferase; AST = aspartate aminotransferase.

inline image

Any adverse event resulting in death, hospitalization, prolongation of hospitalization, or development of a life-threatening or debilitating condition was categorized as a “serious adverse event.” Sixty subjects (31 in the doxycycline group and 29 in the placebo group) reported a total of 82 serious adverse events, none of which was attributable to doxycycline. No deaths occurred during the study.


Our decision to recruit for this trial obese women ages 45–64 years who had radiographic evidence of unilateral knee OA on the standing AP radiograph was based on data from a population-based study indicating that 47% of women with these characteristics developed radiographic evidence of incident OA in the contralateral knee 24 months later (10). It should be noted, however, that the incidence of OA in that study was based primarily on osteophytosis; JSN was factored into the grading of OA progression, but was scored semiquantitatively with the use of a validated atlas and was not measured quantitatively, as in the present study.

Therefore, when this trial was implemented, we anticipated that we would be able to evaluate the effects of doxycycline on the progression of established OA in the index knee and on incident OA in the contralateral knee. However, during the course of the study, we found that although the contralateral knee is radiographically normal in a conventional standing AP view, in most cases it exhibits evidence of OA in a lateral, semiflexed AP and/or patellofemoral view (33). Hence, this study does not assess the effect of doxycycline on incident OA in the contralateral knee but rather on the progression of relatively mild OA in that joint, which was not apparent in the baseline standing AP view.

While the primary outcome variable in this trial was JSN in the medial tibiofemoral compartment, it should be noted that the eligibility criteria did not preclude enrollment of subjects whose index knee showed JSN in the lateral compartment at baseline. In fact, 10% of index knees exhibited isolated lateral JSN in the baseline semiflexed AP view. Lateral JSN in the index knee progressed in 7% of subjects in the doxycycline group (and in 4% of subjects in the placebo group) and did not confound our ability to detect a significant effect of doxycycline on medial JSN in the index knee. In addition, 5% of subjects in the doxycycline group and 4% of subjects in the placebo group exhibited progression of lateral JSN in the contralateral knee, in which there was no indication at baseline of involvement of the lateral compartment. Therefore, the failure of doxycycline to slow the mean rate of medial JSN in the contralateral knee was not due to lateral compartment narrowing with concomitant widening of medial compartment JSW.

After oral administration of a single 100-mg dose of doxycycline, the serum half-life of the drug in normal subjects is ∼16 hours (34). Selection of the dose used in the present study, 100 mg twice a day, was based on findings in humans with OA undergoing joint replacement surgery: when administered for 5 days prior to surgery, doxycycline at this dose significantly reduced the levels of total and active collagenase and gelatinase in extracts of the OA cartilage, whereas this effect was not achieved with other dosing regimens (7). In vitro, 10 μM and 30 μM doxycycline inhibited cartilage gelatinase activity by 44% and 82%, respectively (1). Degradation of embryonic avian tibias was inhibited by exposure in vitro to a similarly low concentration of the drug, following which collagenase and 62-kd gelatinase were no longer detectable in the spent culture medium (35).

The feasibility of the protocol employed in this trial is evidenced by the fact that 30-month radiographs were obtained for 85% of all subjects randomly assigned to treatment and by the high mean rate of adherence to the dosing regimen (∼90% among completers and >80% among all subjects for whom 30-month radiographic data were available). Furthermore, 71% of all patients randomly assigned to treatment were still taking doxycycline or placebo 30 months later. The approaches we used to optimize adherence to the dosing regimen and minimize the dropout rate are described in detail elsewhere (36).

It was an advantage for this trial that doxycycline is generally well tolerated and has a long track record of relative safety and well-recognized adverse effects (e.g., sun sensitivity, monilial vaginitis, nonspecific GI symptoms). Only these 3 types of adverse effects occurred significantly more frequently in the active treatment group than in the controls, and no serious adverse events were attributed to doxycycline. On the one hand, the fact that subjects were permitted to take prescription and over-the-counter NSAIDs/analgesics throughout the trial (except during washout periods preceding clinical evaluations) presumably also contributed to the low dropout rate and high level of adherence to the dosing regimen. On the other hand, continued use of NSAIDs/analgesics undoubtedly contributed to some of the nonspecific GI symptoms and to the dropout rate in both treatment groups. Although the emergence of features of the MMP-inhibitor syndrome (e.g., shoulder periarthritis, Dupuytren's contracture) (37) was not actively sought, these were not detected by the adverse event reporting mechanism employed.

In both knees and both treatment groups, the mean rate of JSN at 30 months was about twice as great as that at 16 months (Table 2). In the index knee, doxycycline reduced the mean rate of JSN at 30 months by 33%; notably, a clear difference between doxycycline and placebo was already apparent in the 16-month data (40% less rapid JSN in the doxycycline group than in the placebo group; P = 0.027). However, the drug had no effect on progression of the K/L grade in either knee.

At 16 months, the rate of JSN in the contralateral knee was identical in the 2 treatment groups; at 30 months, the rate was numerically greater in the doxycycline group than in the placebo group, but the difference was not statistically significant (Table 2). Although the possibility cannot be excluded that the difference might have reached significance if the sample size had been larger, the lack of evidence that doxycycline slowed the rate of JSN in the contralateral knee may have been due to the fact that some matrix degradation pathways are more important in the later stages of OA than in the earlier stages. Doxycycline may have interfered with processes driving cartilage breakdown in the index knee that were not (yet) operative in the contralateral knee.

In support of this possibility, synthesis of type II collagen and turnover of aggrecan were found to be increased in overt focal lesions in femoral articular cartilage of older humans, thus resembling changes in established OA, but no up-regulation of synthesis accompanied earlier lesions (38). Messenger RNA for type II and type III collagen, biglycan, and MMPs 2, 11, and 13 was detected only in cartilage from patients with advanced OA, while the gene for MMP-3 was up-regulated only in samples of cartilage with structural evidence of early OA, but was undetectable in cartilage with more advanced disease (39).

It should be noted, however, that doxycycline had a protective effect in animal models of OA when given prophylactically (6, 8), suggesting that in these models it is active in the very early stages of OA. The difference between the effects of doxycycline on early OA in animals and in humans may be due to differences among species in the relative contributions of various mediators or enzymes to articular cartilage pathology (40).

Doxycycline treatment did not result in a change in clinical outcomes that fulfilled recently proposed responder criteria for OA (41). Those criteria, however, were evaluated in OA patients in whom the baseline level of symptoms was relatively high, and they have not been adequately tested in patients with less severe symptoms, as in the present study. Even though doxycycline did not alter mean scores for joint pain, it significantly decreased the frequency with which subjects reported ≥20% increases in knee pain in successive semiannual visits. Our requirement that the increase in pain on a 10-cm VAS be ≥1 cm in magnitude corresponds closely to the definition of a minimum clinically important difference in patients with OA (42). Notably, the effect of doxycycline on symptoms was not as apparent in the contralateral knee, in which the drug had no effect on JSN and no clinical benefit.

In both knees, the rate of JSN was more than twice as rapid in subjects who reported frequent increases in knee pain as in those with a stable pain score, appearing to validate the clinical importance of retardation of articular cartilage loss. Joint pain may possibly serve as an indicator of synovitis that leads to cartilage destruction. Clinically unrecognized synovitis has been detected arthroscopically in up to one-third of patients with radiographic OA. Sites of synovial inflammation may abut cartilage lesions (43), suggesting that synovitis may cause localized chondropathy. On the other hand, destruction of joint cartilage, perhaps driven by the abnormal mechanical environment of the OA joint, may be the cause of synovitis.

Finally, some limitations of this study should be noted. The sample size was relatively small and the duration of followup was relatively brief, considering that changes in joint pain, function, and structure in patients with knee OA typically occur over many years. Furthermore, our patients were highly selected with respect to sex, BMI, age, and discordance in the severity of OA pathology in their two knees, limiting generalizability of the study results. It cannot be assumed that the protective effect of doxycycline noted in the index knee would occur at other joint sites (e.g., hips or hands), in subjects of all age groups, or in men. Finally, given the higher rate of adverse reactions in the doxycycline group than in the placebo group, the subjects may not have been completely blinded to their treatment assignments. However, although it is possible that unblinding, if it did occur, may have had some influence on symptomatic outcomes, it is unlikely to have affected the progression of structural damage in the OA joint.


We are grateful to the dedicated group of Study Coordinators whose skills were essential in assuring the successful conduct of this trial: Autumn Bill, Cherita Fagan, Nancy Flowers, Johnny Fuessler, Dot Gunter, Jenny Hixon, Julie Johnson, Sherrie Pryber, Linda Ragozzino, Andrea Schaffter, Edie Sparr, and Kathy Urbansky. Beverly Musick provided excellent help with data management and coordination. We thank Professor Michel Lequesne for his assistance with manual measurements of a subset of the knee radiographs to determine interreader reproducibility of measurements of JSW and of the magnification marker. Mark Schluchter was particularly helpful in the evaluation of missing data. Niall Doherty and Thasia Woodworth (Pfizer, Inc.) were of great assistance in obtaining doxycycline and matched placebo. Kathie Lane provided invaluable clerical assistance.