Hide

Article Summarizer

Summarize any article into our unique interactive flashcard format, and accelerate your reading and learning with our DSA-approved technology. Perfect also for screening, review and research communication.

How the weather affects the pain of citizen scientists using a smartphone app

This study has demonstrated that higher relative humidity and wind speed, and lower atmospheric pressure, were associated with increased pain severity in people with long-term pain conditions

Summary Flashcard

Key concepts

Abstract

Patients with chronic pain commonly believe their pain is related to the weather. Scientific evidence to support their beliefs is inconclusive, in part due to difficulties in getting a large dataset of patients frequently recording their pain symptoms during a variety of weather conditions. Smartphones allow the opportunity to collect data to overcome these difficulties. Our study Cloudy with a Chance of Pain analysed daily data from 2658 patients collected over a 15-month period. The analysis demonstrated significant yet modest relationships between pain and relative humidity, pressure and wind speed, with correlations remaining even when accounting for mood and physical activity. This research highlights how citizen-science experiments can collect large datasets on real-world populations to address long-standing health questions. These results will act as a starting point for a future system for patients to better manage their health through pain forecasts.

Synopsis

Scholarcy highlights

Scholarcy summary

Introduction

Weather has been thought to affect symptoms in patients with chronic disease since the time of Hippocrates over 2000 years ago.[1].

Studies have failed to reach consensus in part due to their small sample sizes or short durations; by considering a limited range of weather conditions; and heterogeneity in study design.[5,6,7,8,9,10,11] Resolving this question requires collection of high-quality symptom and weather data on large numbers of individuals

Such data need to include other factors potentially linked to daily pain variation and weather, such as mood and amount of physical activity.

Collecting this kind of multi-faceted data in large populations over long periods of time, has been difficult

Methods

Patient involvement

Patient involvement has been important throughout the study, from inception to interpretation of the results.

C.G. and other members of the Patient and Public Involvement Group were involved in media broadcasts at study launch and subsequent public engagement activities, explaining why the research question was important to them and relevant to patients with long-term pain conditions.[22] They have supported the interpretation of findings and the development of dissemination plans for the results, ensuring the results reach study participants, patient organizations and the general public.

Results

Recruitment and retention

The study app was downloaded by 13,207 users over the 12-month recruitment period (Figs 1 and 2a) with recruitment from all 124 UK postcode areas.

A total of 10,584 participants had complete baseline information and at least one pain entry, with 6850 (65%) participants remaining in the study beyond their first week and 4692 (44%) beyond their first month (Fig. 2b).

A total of 2658 participants had at least one hazard period matched to a control period in the same month (Fig. 3) and were included in the final analysis.

There were 9695 hazard periods included in the analysis for the final 2658 participants, matched to 81,727 control periods in 6431 participant-months.

A total of 1235 participants contributed one month, and the remaining 1423 participants contributed 2–15 months

Conclusion

This study has demonstrated that higher relative humidity and wind speed, and lower atmospheric pressure, were associated with increased pain severity in people with long-term pain conditions.

The most significant contribution was from relative humidity.

The ‘worst’ combination of weather variables would increase the odds of a pain event by just over 20% compared to an average day.

Such an increased risk may be meaningful to people living with chronic pain

Comparative analysis

Builds on previous research
Nonetheless, such an increased risk may be meaningful to people living with chronic pain. In addition to investigating the weather–pain relationship, we successfully conducted a national smartphone study that delivered on the promise of how consumer technology can support health research.[12,15] This study recruited over 10,000 participants throughout the United Kingdom, sustained daily self-reported data over many months,[13] and showcased the value of passively collected GPS data
Differs from previous work
In addition to investigating the weather–pain relationship, we successfully conducted a national smartphone study that delivered on the promise of how consumer technology can support health research.[12,15] This study recruited over 10,000 participants throughout the United Kingdom, sustained daily self-reported data over many months,[13] and showcased the value of passively collected GPS data. Prior large smartphone studies have retained only around one in ten participants for seven days or less.[16,17] In contrast, our study retained 65% of participants for the first seven days, and 44% for the first month, with over 2600 participants contributing to the analysis having provided data for many months of the study.[13,14] An important success factor was strong public involvement in early setup and piloting, as well as participants’ interest in weather as a possible pain trigger.[14]
Builds on previous research
It is possible that only people with a strong belief in a weather–pain relationship participated, generating an unrepresentative sample. However, the percentage of participants who believed in the weather–pain relationship was similar to prior studies,[20] and we did not see selective attrition of people who reported no weather–pain beliefs.[13]
Builds on previous research
C.G. and other members of the Patient and Public Involvement Group were involved in media broadcasts at study launch and subsequent public engagement activities, explaining why the research question was important to them and relevant to patients with long-term pain conditions.[22]. They have supported the interpretation of findings and the development of dissemination plans for the results, ensuring the results reach study participants, patient organizations and the general public

Figures

Figure 1: User interface of the study app (uMotif, London). Each colored segment represents one of the ten data items. Participants report their symptoms on a five-point scale by dragging the segment from the center outwards predominantly female (83%), had a mean age of 51 years (standard deviation 12.6), and had a range of different pain conditions, predominantly arthritis (Supplementary Table 1). The median number of weather stations associated with each participant during the course of their active data-collection period was 9 (IQR 4–14) with a maximum of 82 stations, indicating how mobile participants were during the course of the study and the importance of accounting for the weather at different locations over the course of the study. As an illustration of the structure of the data, the proportion of participants reporting a pain event was plotted as a heat map per calendar day for the study period (Fig. 4), aligned with the average United Kingdom weather data for the same time period. On any given day during the study, about 1–6% of participants had a pain event. At the start of the study, most participants believed in an association between weather and their pain (median score 8 out of 10, IQR 6–9). The demographics, health conditions and baseline beliefs of the 2658 participants included in the analysis were representative of the 10,584 participants who downloaded the app and provided baseline information (Supplementary Table 2)
Figure 2: Recruitment and retention. a Cumulative recruitment and number of active participants through time. The blue line represents the cumulative number of participants with a completed baseline questionnaire and at least one pain score submitted. The red line represents the current number of active participants (i.e. those who have submitted their first but not yet their last pain score in the study period). b Retention through time. The graph represents the retention of active participants through time as a survival probability from the day of their recruitment. Participants were censored when they were no longer eligible for follow-up. Eligible follow-up time ranged from 90 days (for those recruited on 20 January 2017) to 456 days (for those recruited on 20 January 2016)
Figure 3: Example participant timeline of 21 days, showing participant-reported items (here, pain severity, mood, and exercise) and weather data (here, temperature and relative humidity). Pain events with their associated hazard periods (dark grey) occur when pain severity increases by two or more ordinal categories between consecutive days (e.g. from Day 4 to Day 5). Control periods (light gray) occur on days that were eligible to be a pain event, but where pain did not increase by two or more ordinal categories. Days where there was no recorded pain on the preceding day, or where the preceding day’s pain was severe or very severe (and could thus not increase by two or more categories), were not eligible to be pain-event days or control days. The case-crossover analysis compared the weather on pain-event days to weather on control days within a risk set of a calendar month only around one in ten participants for seven days or less.[16,17] In contrast, our study retained 65% of participants for the first seven days, and 44% for the first month, with over 2600 participants contributing to the analysis having provided data for many months of the study.[13,14] An important success factor was strong public involvement in early setup and piloting, as well as participants’ interest in weather as a possible pain trigger.[14] The study design has resolved problems of prior weather–pain studies such as small populations,[5,7] short follow-up,[3,8] surrogate pain outcomes,[11] the absence of possible causal pathway variables such as mood, and assumptions about where participants were located and thus the weather to which they were exposed.[18,19]
Figure 4: The proportion of eligible active participants reporting a pain event during the study period, aligned with average UK weather data from February 2016 to April 2017. Heat map colors indicate the percentage of participants reporting a pain event on that day, ranging from 1–6% participants. The denominator per day is the number of participants who reported their pain on the day of interest and the prior day, irrespective of the level of pain on the prior day and thus their eligibility for a pain event someone else’s “severe”. The within-person case-crossover analysis meant we compared moments when an individual’s score increased by a meaningful amount to a control period for that same person. Fifth, we chose to model the weather using daily averages. It is possible that other findings may be hidden if the association between weather and pain was with other metrics of weather, such as the daily maximum, minimum, or range, or even if the changes in weather on hourly time scales affect participants’ pain. Sixth, the findings from this United Kingdom study cannot necessarily be extrapolated to different climates where the weather is different. Seventh, our population-wide analysis assumed that all participants have the same weather–pain relationship. Different diseases may have different sensitivities to pain and, even within disease, participants may be affected differently. Our decision to use the whole chronic-pain population in our primary analysis means the overall associations with weather variables may be combinations of strong, weak and absent causal effects, thereby underestimating the most important associations. Notable differences were not seen after stratification by pain condition, although the power to detect any differences was reduced because of smaller sample sizes. Lastly, the inclusion of repeated events per person required us to consider within-subject dependence which, if not accounted for, would lead to bias.[21] Our outcome was based on changes in pain (a two or more category increase), which meant events rarely occurred on consecutive days, thereby ensuring a time gap between recurrent events and the avoidance of bias
Figure 5: Estimated odds of a painful day for all weather days experienced during the 15 months. Estimated odds of a painful day are plotted as the odds ratio for each day compared to the average weather day in this period (temperature = 9.3 °C, relative humidity = 83%, wind speed = 4 m s–1 and pressure = 1013 mbar). Estimated odds are calculated from the output of the multivariable regression analysis. The day associated with the highest estimated odds of a pain event had a temperature of 9 °C, relative humidity 88%, wind speed 9.5 m s–1 and pressure 988 mbar. The day associated with the lowest estimated odds of a pain event was when the temperature was 7 °C, relative humidity was 67%, wind speed 4.5 m s−1 and pressure 1030 mbar

Methods

Patient involvement

Patient involvement has been important throughout the study, from inception to interpretation of the results. Co-author C.G. is a patient partner and co-applicant, while a patient and public involvement group of seven additional members has supported the study, meeting eight times in total. During the feasibility study,[14] patients positively influenced the wording and display of questions within the app. C.G. and other members of the Patient and Public Involvement Group were involved in media broadcasts at study launch and subsequent public engagement activities, explaining why the research question was important to them and relevant to patients with long-term pain conditions.[22] They have supported the interpretation of findings and the development of dissemination plans for the results, ensuring the results reach study participants, patient organizations and the general public.

We recruited participants through local and national media (television, radio, and press) and social media from 20 January 2016 to 20 January 2017. To participate in the study, participants needed to (i) be living with long-term (>3 months) pain conditions, (ii) be aged 17 years or older, (iii) be living in the United Kingdom, and (iv) own an Android or Apple iOS smartphone. Interested participants were directed to the study website (www.cloudywithachanceofpain.com) where they could check their eligibility, learn about the study, and download the uMotif app (Fig. 1). After downloading the study app, participants completed an electronic consent form and a baseline questionnaire including demographic information (sex, year of birth, first half of postcode), anatomical site(s) of pain, underlying pain condition(s), baseline medication use, and beliefs about the extent to which weather influenced their pain on a scale of 0–10, including which weather condition(s) were thought to be most associated with pain. Participants were then invited to collect daily symptoms for six months, or longer if willing. Each day, the app alerted participants to complete ten items at 6:24 p.m. (Fig. 1). The ten items were pain severity, fatigue, morning stiffness, impact of pain, sleep quality, time spent outside, waking up feeling tired, physical activity, mood, and well-being. Each data item had five possible labeled ordinal responses. For example, in response to the question “How severe was your pain today?”, possible responses were “no pain”, “mild pain”, “moderate pain”, “severe pain” or “very severe pain”. The data were analysed using a case-crossover design where, for each participant, exposure during days with a pain event (“hazard periods”) were compared to “control periods” without a pain event in the same month.[23] Pain events were defined as a two-or-more category increase in pain from the preceding day, consistent with more stringent definitions of a clinically important difference[24] (Fig. 3). Data collection ended on 20 April 2017.

Limitations

There are potential limitations to this study. First, the reduction in participant numbers from over 10,000 with baseline data to the final 2658 participants with at least one within-month risk set raises questions about generalisability. Importantly, the characteristics of those included in the analysis were similar to the initial 10,000 participants, other than being slightly older (mean age 51 versus 48 years old). In a prior analysis, we showed that Cloudy participants were largely representative of a population reporting chronic-pain symptoms,[13] although proportionally fewer participants at both extremes of age were recruited. However, we would not expect middle-aged recruits to differ in their relationship between weather and pain from older or younger participants, and thus such selection factors would not invalidate our results. Second, the study was advertised to participants with a clear research question. It is possible that only people with a strong belief in a weather–pain relationship participated, generating an unrepresentative sample. However, the percentage of participants who believed in the weather–pain relationship was similar to prior studies,[20] and we did not see selective attrition of people who reported no weather–pain beliefs.[13] The within-person design would, regardless, mean that participants who drop out early would not introduce bias from time-invariant characteristics. Third, the lack of blinding raises possible information bias where observed weather could influence participants’ symptom reporting. Our baseline questionnaire demonstrated that rain and cold weather were the most common pre-existing beliefs. If a reporting bias were to exist, we would expect higher pain to be reported at times of colder weather. Our findings—including the absence of an association with either temperature or rainfall—cannot be explained by such a reporting bias. Fourth, pain reporting is subjective, meaning one participant’s “moderate” might equate to someone else’s “severe”. The within-person case-crossover analysis meant we compared moments when an individual’s score increased by a meaningful amount to a control period for that same person. Fifth, we chose to model the weather using daily averages. It is possible that other findings may be hidden if the association between weather and pain was with other metrics of weather, such as the daily maximum, minimum, or range, or even if the changes in weather on hourly time scales affect participants’ pain. Sixth, the findings from this United Kingdom study cannot necessarily be extrapolated to different climates where the weather is different. Seventh, our population-wide analysis assumed that all participants have the same weather–pain relationship. Different diseases may have different sensitivities to pain and, even within disease, participants may be affected differently. Our decision to use the whole chronic-pain population in our primary analysis means the overall associations with weather variables may be combinations of strong, weak and absent causal effects, thereby underestimating the most important associations. Notable differences were not seen after stratification by pain condition, although the power to detect any differences was reduced because of smaller sample sizes. Lastly, the inclusion of repeated events per person required us to consider within-subject dependence which, if not accounted for, would lead to bias.[21] Our outcome was based on changes in pain (a two or more category increase), which meant events rarely occurred on consecutive days, thereby ensuring a time gap between recurrent events and the avoidance of bias.

References Download

1. Hippocrates. On Airs, Waters and Places. http://classics.mit.edu//Hippocrates/airwatpl.html. (400 B.C.E.).
2. Hagglund, K. J. et al. Weather, beliefs about weather, and disease severity among patients with fibromyalgia. Arthritis Rheum. 7, 130–135 (1994).
3. Timmermans, E. J. et al. Self-perceived weather sensitivity and joint pain in older people with osteoarthritis in six European countries: Results from the European Project on OSteoArthritis (EPOSA). BMC Musculoskelet. Disord., https://doi.org/10.1186/1471-2474-15-66 (2014).
4. Smedslund, G. & Hagen, K. B. Does rain really cause pain? A systematic review of the associations between weather factors and severity of pain in people with rheumatoid arthritis. Eur. J. Pain. 15, 5–10 (2011).
5. Aikman, H. The association between arthritis and the weather. Int. J. Biometeorol. 40, 192–199 (1997).
6. Brennan, S. A. et al. Influence of weather variables on pain severity in end-stage osteoarthritis. Int. Orthop. 36, 643–646 (2012).
7. Smedslund, G. et al. Does the weather really matter? A cohort study of influences of weather and solar conditions on daily variations of joint pain in patients with rheumatoid arthritis. Arthritis Rheum. 61, 1243–1247 (2009).
8. Bossema, E. R. et al. Influence of weather on daily symptoms of pain and fatigue in female patients with fibromyalgia: a multilevel regression analysis. Arthritis Care Res. (Hoboken) 65, 1019–1025 (2013).
9. Duong, V. et al. Does weather affect daily pain intensity levels in patients with acute low back pain? A prospective cohort study. Rheumatol. Int. 36, 679–684 (2016).
10. Guedj, D. & Weinberger, A. Effect of weather conditions on rheumatic patients. Ann. Rheum. Dis. 49, 158–159 (1990).
11. Jena, A. B. et al. Association between rainfall and diagnoses of joint or back pain: retrospective claims analysis. BMJ 359, j5326 (2017).
12. Hayden, E. C. Mobile-phone health apps deliver data bounty. Nature 531, 422–423 (2016).
13. Druce, K. L. et al. Recruitment and ongoing engagement in a UK smartphone study examining the association between weather and pain: Cohort study. JMIR mHealth uHealth 5, e168 (2017).
14. Reade, S. et al. Cloudy with a Chance of Pain: Engagement and subsequent attrition of daily data entry in a smartphone pilot study tracking weather, disease severity, and physical activity in patients with rheumatoid arthritis. JMIR mHealth uHealth 5, e37 (2017).
15. Santos-Lozano, A. et al. mHealth and the legacy of John Snow. Lancet 391, 1479–1480 (2018).
16. Bot, B. M. et al. The mPower study, Parkinson disease mobile data collected using ResearchKit. Sci. Data 3, 160011 (2016).
17. McConnell, M. V. et al. Feasibility of obtaining measures of lifestyle from a smartphone app. JAMA Cardiol. 2, 67 (2017).
18. Smedslund, G. et al. Do weather changes influence pain levels in women with fibromyalgia, and can psychosocial variables moderate these influences? Int. J. Biometeorol. 58, 1451–1457 (2014).
19. Beilken, K. et al. Acute low back pain? Do not blame the weather—A case-crossover study. Pain. Med. 18, pnw126 (2016).
20. Jamison, R. N., Anderson, K. O. & Slater, M. A. Weather changes and pain: perceived influence of local climate on pain complaint in chronic pain patients. Pain 61, 309–315 (1995).
21. Luo, X. & Sorock, G. S. Analysis of recurrent event data under the case-crossover design with applications to elderly falls. Stat. Med. 27, 2890–2901 (2008).
22. Cloudy with a Chance of Pain on BBC North West Tonight. https://www.youtube.com/watch?v=YUdtKGr49GY. Accessed 14 Oct 2019 (2016).
23. Maclure, M. The case-crossover design: a method for studying transient effects on the risk of acute events. Am. J. Epidemiol. 133, 144–153 (1991).
24. Olsen, M. F. et al. Pain relief that matters to patients: systematic review of empirical studies assessing the minimum clinically important difference in acute pain. BMC Med. 15, 35. https://doi.org/10.1186/s12916-016-0775-3 (2017).
25. BPH postcodes. A brief guide to UK postcodes. https://www.bph-postcodes.co.uk/guidetopc.cgi (2018).
26. Isaacman, S., Becker, R., Martonosi, M., Rowland, J., & Varshavsky, A. Identifying important places in people’s lives from cellular network data sibren. Proc. of 9th Int. Conf. Pervasive Comput. 1–18 (2011).
27. Hartigan, J. A. Clustering Algorithms. (Wiley, New York, 1975).
28. Mittleman, M. A. & Mostofsky, E. Exchangeability in the case-crossover design. Int. J. Epidemiol. 43, 1645–1655 (2014).
29. Burnham, K. P., Anderson, D. R. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. 2nd ed. (Springer-Verlag, New York, 2003).
30. R Core Team. R: A language and environment for statistical computing, https://www.r-project.org (2018).