If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Requests for reprints should be addressed to Timothy McAlindon, MD, MPH, Division of Rheumatology, Tufts-New England Medical Center, Box 406, 750 Washington Street, Boston, MA 02111.
Individuals with osteoarthritis often assert that change in the weather influences their pain, but the evidence is inconclusive. Our objective was to determine if short-term weather parameters influence knee osteoarthritis pain.
Methods
We performed a longitudinal analysis of pain reports from a 3-month clinical trial among individuals with knee osteoarthritis dispersed across the United States. Daily values for temperature, barometric pressure, dew point, precipitation, and relative humidity were obtained from the weather station closest to each participant. We used a longitudinal mixed-model random effects analysis with a first-order autoregressive error structure to test for associations while accounting for within-patient correlation.
Results
The study included 200 participants with knee osteoarthritis. Their mean age was 60 years (standard deviation [SD] 9.4), 64% were female, and 10.5% were African American or Hispanic. They had a mean body mass index of 32.5 kg/m2 (SD 8.4) and a baseline WOMAC pain score of 9.0 (SD 3.4). There were consistent associations of pressure change and ambient temperature with pain severity (change in barometric pressure, coefficient = 1.14, P = .02, ambient temperature = −0.01, P = .004; adjusted mutually and for age, gender, body mass index, nonsteroidal anti-inflammatory drug use, opiate use, and prior pain score). Interaction terms between change in barometric pressure and ambient temperature had no influence in the models.
Conclusions
Changes in barometric pressure and ambient temperature are independently associated with osteoarthritis knee pain severity.
However, a consistent relationship between joint pain and weather factors has been strikingly difficult to prove. For example, a recent systematic review of 16 studies of joint pain and weather found no consensus on the issue.
On the other hand, studies of this question face serious methodologic obstacles inherent to evaluating a relationship between a subjective outcome and a universally apparent exposure, especially in the face of firmly held convictions. As pointed out, these limitations are widely represented among the body of current evidence.
Particular problems include disclosure of the study hypothesis to participants, small numbers of participants, and study samples heterogeneous with respect to rheumatic disorders and observation periods.
Knee pain severity in individuals with knee osteoarthritis is modestly influenced by the weather.
•
An increase in barometric pressure is associated with greater pain.
•
Colder ambient temperature is associated with greater pain.
A completed Internet-based clinical trial among 205 participants with confirmed knee osteoarthritis provided a unique opportunity to investigate a relationship between arthritis pain and meteorologic conditions in a setting that avoids many of these inherent problems.
The participants in that trial were geographically dispersed within the United States and participated at different times of the year. The topic of weather as a research question (for participants or investigators) arose only after completion of the trial, eliminating the opportunity for bias because of disclosure of the study hypothesis. In addition, we obtained meteorologic data through an entirely separate prospective collection mechanism.
Finally, most reports of weather influences suggest that these effects operate in the short term.
Therefore, we set out to test relationships of pain not only with ambient conditions but also with change in conditions immediately preceding each pain report.
Methods
Sample: The Online Glucosamine Trial
The Online Glucosamine Trial was performed between March 2000 and May 2003. We previously reported the methodology and results of this trial in detail.
Briefly, this was an online deployment of a rigorous randomized placebo-controlled 3-month trial of glucosamine sulfate for knee osteoarthritis symptoms. The participants were individuals with knee osteoarthritis classified according to American College of Rheumatology criteria.
The adherence rate for all scheduled visits was 77%. The outcome of the trial itself was entirely negative, with no difference in pain severity found between the 2 groups at any time point.
Participants in the Online Glucosamine Trial who were also eligible for the weather analysis were those for whom we were able to obtain meteorologic data for 7 consecutive days before at least 2 pain reports (n = 200). The study was approved by the institutional review board at Boston University School of Medicine.
Pain Assessments
The primary outcome measure was the pain subscale of the WOMAC questionnaire (Likert version)
administered every 2 weeks over the Internet for a total of 7 assessments. This 5-item inventory provides a score with range 0 to 20 reflecting level of pain experienced during different activities of daily life. Because the statistical approach that we used requires a prior pain report to adjust for the intraparticipant correlation, the associations between pain and climatic parameters were assessed at visits 2 through 7.
Acquisition of Meteorologic Data
We identified the most proximate data-collecting weather station to each participant using the search tool provided by the National Oceanographic and Atmospheric Administration website. Next, we obtained the local daily average values for temperature, barometric pressure, dew point, and precipitation for every participant, for each day of their participation in the trial. We computed the relative humidity from temperature and dew point according to the following formula:
(1)
where T = temperature in degrees Celsius and Td = dew point temperature in degrees Celsius.
Exposure Definitions
We analyzed the exposure parameters in 2 ways: short-term ambient values and change values. The short-term ambient values consisted of the means of the daily averages for temperature, barometric pressure, dew point, precipitation, and relative humidity over the 3-day periods before each pain report. The change in weather parameters was computed as the difference in each meteorologic parameter between the day before each pain report and the day of the pain report itself.
Nonmeteorologic exposures in our analyses included age, gender, body mass index, and use of nonsteroidal anti-inflammatory drugs or opiates at any time in the trial.
Analytic Approach
We used a longitudinal mixed-model analysis to test for associations between meteorologic exposures and knee pain severity, a technique that permits full use of available data while controlling for internal correlations and other covariates. This approach treats each pain report from each participant as a separate observation and adjusts for within-participant correlations and the correlation with the prior pain report. Subjects were treated as random effects so the analysis could be adjusted to each individual’s own pain levels. A first-order autoregressive error structure accounted for within-patient correlation.
Although there was no treatment effect evident in our data, we did observe a time trend in pain scores possibly caused by regression to the mean. We adjusted for this by using a logarithmic covariate function in the regression models reflecting the time elapsed from the trial baseline.
In the multivariate models we adjusted for time from baseline, pain score at the prior pain report, and potential confounders (age, gender, body mass index, nonsteroidal anti-inflammatory drug use, and opiate use). We tested 3 approaches to deal with missing pain scores: use of baseline pain, a linear interpolation, and last pain score carried forward. Comparison of these approaches using the Bayesian Information Criterion indicated that the linear interpolation approach functioned best. Therefore, we used that approach in our models. We explored the independent effects of the ambient and change values for the meteorologic parameters in separate and mutually adjusted models and tested for relationships between the meteorologic exposures using interaction terms. Finally, we tested for interactions among the meteorologic exposures, and between these and other covariates such as age and osteoarthritis severity, using interaction terms in the multivariate models.
Results
Data from 200 participants were eligible for this analysis. Their mean age was 60 years (standard deviation [SD] 9.4), 64% were female, and they had a mean body mass index of 32.5 kg/m2 (SD 8.4) and mean baseline WOMAC pain score of 9.0 (SD 3.4). They provided a total of 935 pain reports, 79% of the total possible. The absent data resulted from missing pain reports (n = 265) or missing weather information (n = 72).
The geographic dispersion of the sample is illustrated in Figure 1. The participants were represented by 114 weather stations, 96% of which were within 1 mile of each participant’s residence. The maximum distance was “within 50 miles” and occurred for 3 participants.
The range and variability of the meteorologic exposures are described in Table 1. Three of these, change in barometric pressure, ambient temperature, and ambient dew point, were associated with pain severity in the multivariate models (Table 2). With adjustment for age, gender, body mass index, nonsteroidal anti-inflammatory drug use, opiate use, and prior pain score, the coefficient was 1.0 (P = .04) for change in barometric pressure, −0.01 (P = .004) for ambient temperature, and −0.01 (P = .02) for ambient dew point. In mutually adjusted models there were persistent significant effects from ambient temperature (coefficient = −0.010, 95% confidence limits −0.017 to −0.003, P = .004) and change in barometric pressure (coefficient = 1.14, 95% confidence limits 0.15 to 2.13, P = .02), but not dew point (which was highly correlated with temperature). Both change in barometric pressure and ambient temperature had similar standardized regression coefficients (0.16 and −0.18, respectively). We did not find any significant interactions between change in barometric pressure, ambient temperature, or dew point using interaction terms entered in the regression models, or between these variables and age or radiographic severity. The use of 1-day averages for ambient temperature generated similar mutually adjusted results (coefficient for ambient temperature −0.010, P = .005; coefficient for change in barometric pressure 1.06, P = .04).
Table 1Distribution of Meteorologic Exposures in the Study Sample
The coefficient indicates the magnitude of change in pain severity score expected from a 1-unit change in the independent variable (eg, the coefficient of 1.0 for change in barometric pressure is commensurate with an increase in knee pain score of 1.0 for each 1 inch of mercury change in barometric pressure).
P Value
Temperature (degrees Fahrenheit)
−0.01
.004
0.01
.3
Barometric pressure (inches mercury)
0.03
.7
1.0
.04
Dew point (degrees Fahrenheit)
−0.01
.02
0.004
.7
Precipitation (inches)
−0.5
.2
−0.1
.7
Relative humidity (%)
0.001
.8
−0.003
.6
Conversion formulas: Centigrade to Fahrenheit: C = (F − 32) × 5/9. Barometric pressure: 1 inch mercury = 3.386 kPa.
Average of within-participant means over the 3-day period before each pain report.
† Change between the day before each pain report and the day of the pain report.
‡ Adjusted for regression to the mean, prior pain score, age, sex, body mass index, nonsteroidal anti-inflammatory drug use, and opiate use.
§ The coefficient indicates the magnitude of change in pain severity score expected from a 1-unit change in the independent variable (eg, the coefficient of 1.0 for change in barometric pressure is commensurate with an increase in knee pain score of 1.0 for each 1 inch of mercury change in barometric pressure).
Our study of 200 people with knee osteoarthritis participating in a nationwide online clinical trial suggests that both change in barometric pressure and ambient temperature influence severity of knee pain. This study presented a unique opportunity to test the meteorologic hypothesis in a way that reduced or eliminated many of the biases present in previous studies of this question. Our participants were geographically dispersed and participated at different times of year, generating greater opportunity for weather exposure variability. Further, the topic of weather as a research question arose only after completion of the trial, minimizing the opportunity for bias resulting from disclosure of the study hypothesis. We used the WOMAC pain subscale, a well-validated instrument,
to prospectively collect knee-pain data. We obtained prospectively collected meteorologic data through an entirely separate mechanism and analyzed them with modern statistical mixed-model methods.
These factors are salient because interpretation of prior studies in this field is obfuscated by their methodologic limitations and problems of inference inherent to testing the relationship of pain to observable meteorologic conditions, especially in the face of strongly held convictions. Quick
identified a large number of such problems among prior studies of joint pain and weather scrutinized in his critical review. These flaws included failures to keep participants uninformed about the topic of the study and weather reports, test effects of changes in the weather, and observe participants for periods sufficient to include seasonal variations. We add that previous studies focused on varying patient populations (among whom weather effects might operate differently), had limited geographic dispersion of participants (limiting weather variability), and used heterogeneous outcome measures. These issues could account for inconclusive and apparently contradictory results among the studies scrutinized by Quick.
The 2 significant exposure variables had similar effects in the models based on their standardized coefficients (0.16 for change in barometric pressure and −0.18 for ambient temperature), but the magnitude of their effects was small in relation to changes in pain that are considered to be clinically significant.
Minimal perceptible clinical improvement with the Western Ontario and McMaster Universities osteoarthritis index questionnaire and global assessments in patients with osteoarthritis.
For example, the effect of ambient temperature was equivalent to an increase in knee pain score of 0.1 for each 10°F decrease in temperature. We found it odd that relatively small effects should be so consistent in our results. This led us to explore whether the importance of this exposure might reside in an indirect effect or interaction, but interaction terms in the multivariate models revealed no evidence of an interaction of ambient temperature with the other meteorologic parameters or with age or disease severity. An alternative explanation may relate to the temperature differences between the outdoor and indoor climate. A tendency to remain indoors, especially during extremes of outdoor temperatures, could limit the opportunity of this exposure to exert an effect. Thus, during cold weather, the effect might only manifest among individuals who venture outdoors. Even if the effect among the individuals who venture outdoors was large, the inherent misclassification consequent on combining their data with those who remained indoors would be to attenuate the aggregated effect computed in the statistical models. Because we did not have information on participants’ activities, we were not able to examine this possibility directly.
Our results also suggest that increasing barometric pressure is associated with increased pain. The interpretation of this association in terms of observable weather is not straightforward, because the relationship between pressure change and weather events is variable. In general, severe deterioration in weather is often accompanied by marked fluctuations in pressure. A sequence of events in which an increase in pressure precedes or accompanies the development of precipitation could explain the phenomenon in which some people with arthritis believe they can predict adverse developments in the weather.
A possible weakness of our study is that we did not have a measurement of affect, so were unable to directly explore psychologic factors (eg, mood) as confounders or mediators of weather effects. However, there are several reasons to doubt that psychologic mechanisms could mediate the association of increasing barometric pressure with increased pain. First, changes in pressure are not directly perceptible. Second, we collected our data prospectively, eliminating the opportunity for participants to make retrospective pain assignments that could be biased by the subsequent appearance of bad weather. Although participants could theoretically have been noting weather predictions, it is more likely that they would have ascribed increasing pain to decreasing pressure, in line with the prevalent folklore. Also, variables more likely to have direct psychologic effects, such as precipitation or humidity, had no significant influences in our data.
Another limitation in our study is that we were unable to adjust for analgesics taken on the day of each pain report. Such analgesics might be expected to attenuate any association with pain severity. As such, we anticipate that this would bias our findings toward the null rather than generate spurious associations.
Methodologic issues and differences in exposure and outcome definitions complicate direct comparison of the results of previous studies in this field with our own. For example, a recent systematic review that used a weighted consensus voting approach favored an association of osteoarthritis pain with high barometric pressure,
compared pain scores and weather conditions among 154 individuals with osteoarthritis participating in a 2-year exercise intervention study in Florida and noted positive associations with days of “rising barometric pressure” in a subset. Strusberg et al
early experiments using a controlled-climate chamber. These studies suggested that a simultaneous increase in humidity accompanied by a decrease in barometric pressure increased pain, swelling, and stiffness caused by arthritis. However, these experiments included only 4 individuals with osteoarthritis, 1 of whom was not “weather sensitive.” The timing of the atmospheric changes, and the short observation periods, may also have limited the external validity of their findings.
The plausibility of our findings is predicated on a biological explanation of how change in barometric pressure and temperature might influence osteoarthritis pain. Notably, there is evidence that barometric pressure contributes to joint integrity. Wingstrand et al,
in a study of cadaveric hips, found the intraarticular pressure to be subatmospheric in normal situations. When the intraarticular pressure was equilibrated with the atmosphere, the hip joints exhibited 8 mm of subluxation without significant traction. This shows that atmospheric pressure has a physical role in stabilizing the hip joint.
Furthermore, they found intraarticular pressures to be elevated in the presence of joint effusion. The direct effect of atmospheric pressure on joint biomechanics could have additional consequences in certain situations, such as joints with effusions or those in which a defect of articular cartilage integrity allows communication between the intracapsular space and the richly innervated subchondral bone and marrow. Such pathology-specific pressure effects might explain the inconsistent results of studies that investigated the influence of weather by using heterogeneous samples.
Joint pains during compression are well recognized among divers, especially saturation divers.
The mechanism is unknown but is conjectured to result from the sudden increase in tissue gas tension surrounding the joints causing fluid shifts and interfering with joint lubrication.
Cold temperatures also might affect joint pain through a number of mechanisms.
Temperature could have direct effects on the compliance of periarticular structures and viscosity of synovial fluid, and indirect effects on inflammatory mediators through influences on capillary permeability.
Our data corroborate the general assertions by people with osteoarthritis that weather conditions influence their pain. Agreement on this issue may enhance physician–patient interactions and help them better understand and manage fluctuations in arthritis pain. Although the therapeutic implications of our findings need to be further developed, they may already help people with osteoarthritis plan their lives. Indeed, Osler’s advice for arthritis sufferers in 1892 may have been partly correct when he asserted that “Many cases are greatly helped by prolonged residence in southern Europe or southern California. Rich patients should always be encouraged to winter in the south and in this way avoid cold, damp weather.”
The Principles and Practice of Medicine: Designed for the Use of Practitioners and Students of Medicine/by William Osler (1892). Special Edition. Classics of Medicine Library,
Birmingham1978
We thank Mike Wankum for his helpful comments and suggestions about the weather analysis, Martin Englund for contributing knowledge on compression arthralgias, and Tanya Doan for the quotation from Osler.
Minimal perceptible clinical improvement with the Western Ontario and McMaster Universities osteoarthritis index questionnaire and global assessments in patients with osteoarthritis.
The Principles and Practice of Medicine: Designed for the Use of Practitioners and Students of Medicine/by William Osler (1892). Special Edition. Classics of Medicine Library,
Birmingham1978
This study was supported initially by a grant from the Arthritis Foundation and subsequently through support from the National Library of Medicine (RO-1 LM06856-01).
Average of within-participant means over the 3-day period before each pain report.
01026-6&id=gr1.jpg){kind=link}