Comparison of Accelerometer-Derived Physical Activity Levels Between Individuals with and Without Cancer: A UK Biobank Study

Abstract

Aim: To identify the difference in physical activity (PA) levels between individuals with and without cancer, and to estimate all-cause mortality associated with this difference. Methods: Current cancer, cancer survivor and cancer-free groups were identified from the UK Biobank. We used multivariate and Cox regression to estimate PA differences and association of PA with all-cause mortality. Results: Compared with the cancer-free individuals, participants in the two cancer groups had fewer minutes in moderate-to-vigorous PA per day in adjusted analyses. The PA difference was associated with higher mortality in the current cancer group. Conclusion: Patients with a history of cancer were less active than those without cancer, and PA is associated with increased mortality. PA improvement strategies in cancer patients must be explored.

Keywords::

• accelerometer
• all-cause mortality
• cancer
• physical activity
• UK Biobank

The number of patients who are living with or have survived cancer in the UK is estimated at around three million [1]. Studies that investigate the relationship between self-reported or accelerometer-derived physical activity (PA) and cancer suggest that PA is an important predictor of outcomes prior to, during and after a cancer diagnosis [2–4], and is correlated with Eastern Cooperative Oncology Group status score, which is an important prognostic factor for patients with cancer [5].

Despite the growing evidence base and number of clinical guidelines describing a positive association of PA with improved outcomes for cancer patients [6–8], cross-sectional data suggest that cancer survivors [9–12] are more likely to be inactive compared with the cancer-free population. Reasons for less PA among the cancer population include reduced pulmonary system efficiency due to tumor burden, anticancer therapy, impairment in heart rate response as a result of chemotherapy or radiation, decrease in red blood cell count [13] and psychological barriers, including difficulty in getting motivated and remaining disciplined [14]. Fatigue [15], tiredness or exhaustion related to cancer and its treatment can persist beyond treatment and is an important contributor to low levels of PA among cancer survivors. The current study aimed to identify differences in PA level among patients with cancer, cancer survivors and a cancer-free population using data from a large prospective UK cohort. The study further investigated the potential impact of the difference between PA levels on all-cause mortality.

Methods

Data source

The UK Biobank is a large [16], population-based prospective study that provides data on the genetic and nongenetic determinants for diseases in middle age and older people in the UK. Demographic and clinical data were collected from 500,000 participants aged 40–69 years in assessment centers across England, Wales and Scotland between 2006 and 2010 (baseline visit). First and second reassessments were carried out between 2012 and 2013, and 2014 and 2017, respectively, each with approximately 20,000 participants. Data from cancer registries were linked to the UK Biobank cohort to provide complete information on cancer diagnoses prior to the start of the study and following recruitment into the study [17] cohort.

Of the total recruited UK Biobank sample, 236,519 participants were asked to join the accelerometer study [18] to obtain objective PA data under free-living conditions. The response rate to the invitations was 44%. Participants were asked to wear an accelerometer device (Axivity AX3 wrist-worn triaxial accelerometer) on the wrist they write with as soon as it was received. The devices were set to turn themselves on and off at predetermined times. The accelerometer captured triaxial acceleration data over a 7-day period at a sampling rate of 100 Hz (1 Hz = 1 sample per second) within a dynamic range of ±8 g (g = standard gravity = 9.81 m/s²). Participants were asked to mail the accelerometers back at the end of 7 days. A total of 103,720 participants returned data between May 2013 and December 2015.

Study population

For this study, we created three cohorts. The current cancer group included participants who had at least one record of a malignant tumor within 5 years before the end of their accelerometer data collection date. In the absence of cancer treatment data to confirm whether individuals were current cancer patients versus cancer survivors, a 5-year cut off was selected based on previous studies that investigated PA among patients with a history of cancer diagnosis; this was to separate short-term from long-term cancer survivors [10,12,19]. Participants who did not have any record in the cancer registry within 5 years before accelerometer data collection end date (regardless of tumor behavior) and who had at least one malignant tumor at least 5 years before the accelerometer data collection were included in the cancer survivor group. The cancer-free group included participants who did not have any record in the cancer registry prior to accelerometer data collection end date.

Participants whose accelerometer data collection end date was after the complete follow-up date for cancer records were excluded. Additionally, data were excluded from participants who did not provide at least 3 days of accelerometer data as well as data in all hours in a 24-h period [18]. Participants with a record of only benign, in situ or uncertain tumors in the cancer registry and no evidence of any malignant tumor were also excluded from the study population. We removed participants with unusually high average time in moderate-to-vigorous physical activity (MVPA) per day, which could indicate accelerometer data error.

Outcomes

We used an open source Python package [18,20] to process the raw accelerometer data and extract the average minutes per day spent in MVPA, which corresponds to PA levels recommended by physical activity guidelines for adults in the UK and the USA [21,22]. Moderate PA is defined as activities that correspond to three to six metabolic equivalents (MET; e.g., walking 4.5 km/h), and vigorous PA includes activities greater than six METs (e.g., running 1.5 km in 10 min), where one MET is the amount of energy spent while at rest [21]. Participants were assumed to be in MVPA if the recorded acceleration was at least 100 milligravities (mgs) averaged over 30-s nonoverlapping time windows, regardless of duration. The threshold of 100 mg is based on a previously published linear regression model that predicts MET levels using accelerometer output, which were developed from wrist-worn accelerometer data collected from 30 adults [23]. An acceleration of 100 mg is predicted to result in an energy expenditure of three METs based on this equation [23]. The 100-mg acceleration threshold for MVPA is common in PA analysis literature [23–25].

We used UK Biobank’s linkage to death registration data [26] to obtain date of death among the current cancer group.

Baseline covariates

Participant demographic and clinical characteristics included age at accelerometer study end date, sex, ethnicity (white, mixed, Asian or Asian British, black or black British, Chinese, other), Townsend deprivation index at the time of recruitment into the UK Biobank, and the following characteristics measured at the time of recruitment into UK Biobank study or during the latest reassessment visit before accelerometer data collection: smoking status (never, previous, current); alcohol intake frequency (≤1–3 times a month, 1–4 times a week, daily or almost daily); BMI (kg/m²; underweight: <18.5, normal: 18.5–25, overweight: 25–30, obese ≥30); exposure to chemical or other fumes (rarely, never, sometimes, often); worked with materials containing asbestos (rarely, never, sometimes, often); worked with paints, thinners or glues (rarely, never, sometimes, often); worked with pesticides (rarely, never, sometimes, often); workplace had a lot of diesel exhaust (rarely, never, sometimes, often); vascular/heart problems diagnosed by doctor (heart attack, angina, stroke, high blood pressure); diabetes diagnosed by doctor; blood clot; deep vein thrombosis; bronchitis; emphysema; asthma, rhinitis; eczema; and allergy diagnosed by doctor. We extracted hospitalization data within the past 3 months of accelerometer data collection among the current cancer group using the hospitalization data linkage of UK Biobank [26]; hospitalization in this period was assumed to be a proxy for cancer severity, which was controlled for in the survival analysis in this group.

Statistical methods

We reported the number and percentage for categorical values and descriptive statistics (mean, standard deviation, median, 25th and 75th percentiles) for continuous variables separately for each of the groups. We compared baseline characteristics and average minutes spent in MVPA per day between the current cancer and cancer-free group, and between the cancer survivor and cancer-free group. For categorical variables, χ² tests were used. For continuous variables, normality of the distributions was assessed using Kolmogorov–Smirnov tests. For normally distributed variables, t-tests were used; for non-normally distributed variables, Wilcoxon Mann–Whitney tests were used to compare means between the groups. p < 0.05 was considered statistically significant.

Separate linear regression models were constructed to compare the difference in MVPA per day between the current cancer and cancer-free group, and between the cancer survivor and cancer-free group. First, univariable regression models were constructed that included all baseline covariates. Key variables known to be important prognostic factors (age, sex, BMI, smoking, alcohol intake) were considered a priori confounders and carried forward to the multivariable model – the rest of the covariates were dropped if they were not significant in the univariable models. In the second step, significant predictors were combined in a multivariable regression model, which used a backward stepwise selection to exclude those that became nonsignificant. We did not impute incomplete or missing data from baseline covariates and cancer status, and participants with missing values in any of the baseline covariates that were included in the multivariable regression were excluded from this analysis. We also dropped variables with very high levels of missingness from the multivariable analyses. For all univariable and final selected multivariable models, coefficient values, their 95% CIs and p-values are reported.

We also investigated the impact of the uncertainty around a patient’s cancer status (current vs survivor) by repeating the comparison of these two groups with the cancer-free cohort by setting the cut-off value for years since last cancer diagnosis to four, three, two and one.

To assess the potential impact of differences in time in MVPA, we modeled all-cause mortality among current cancer patients. We followed participants in the current cancer group from the end of their accelerometer data collection to their date of death as provided by UK Biobank’s linkage to death registration data or to the complete follow-up date for mortality data if they did not have a death record. We used Cox proportional hazard regression to model time-to-death as a function of average hours spent in MVPA per day and controlled for the covariates selected for the multivariable linear regression models. As a sensitivity analysis, we included a binary variable for being hospitalized within the previous 3 months of the accelerometer study. We then reported the hazard ratio and corresponding 95% CI for all-cause mortality associated with an increase in time in MVPA equal to the adjusted difference of time in MVPA between participants in the current cancer group and the cancer-free group.

Results

A total of 103,695 participants participated in the accelerometer study, and 96,699 provided at least 3 days of data, including during each 1-h window in a 24-h period. Among participants with enough accelerometer data, 96,359 completed the study before the complete follow-up dates of the linked cancer registry data. Two accelerometer files were erroneous and another had an average daily MVPA of 964 min (16 h), which indicated a potential measurement error, and they were removed. Out of the remaining 96,356 participants 2596 only had tumors classified as benign, in situ, or uncertain type, and were removed. This left 93,760 participants for the analyses. Among these participants, 4761 had at least one malignant tumor within the last 5 years of their accelerometer data collection (current cancer group), 6419 had no tumors within the last 5 years of accelerometer data collection but at least one malignant tumor before that date (cancer survivors), and 82,580 participants had no records in the cancer registry before their accelerometer data collection end date (cancer-free group) (Figure 1).

Figure 1. Participant flow chart.

*Good wear time criteria are based on [18].

^†Toxicity related variables were removed from the multivariate analysis due to high levels of missing data. Please refer to the Supplementary Material for details on missing data.

Patients in the current cancer and cancer survivor groups were older compared with the cancer-free group, more likely to be previous smokers, slightly more likely to have reported drinking alcohol every day or almost every day at their latest assessment visit, and had higher levels of deprivation at baseline visit. While a smaller percentage of participants in the current cancer group versus the cancer-free group was females, the proportion of females was higher in the cancer survivor group compared with the cancer-free group. Most participants in all three groups were white, although this percentage was slightly lower in the cancer-free group. Participants in both cancer groups had lower PA levels compared with the cancer-free group. Baseline characteristics and PA levels of study participants are summarized in Table 1. A summary of variables related to exposure to hazardous materials is provided in Supplementary Table 1.

Table 1. Demographic characteristics and physical activity levels.

PA level and baseline characteristics	Cancer free (n = 82,580)	Cancer current (n = 4761)	Cancer survivor (n = 6419)
Average daily minutes spent on MVPA; mean (SD) (25th, 50th, 75th percentile)	101.1 (51.7) (57.6, 100.8, 129.6)	87.8^¶ (47.4) (57.6, 86.4, 115.2)	90.3^¶ (48.5) (57.6, 86.4, 115.2)
Age at accelerometer study end date (years); mean (SD) (25th, 50th, 75th percentile)	61.9 (7.9) (55.6, 62.8, 68.1)	66^¶ (6.8) (62.4, 67.4, 71.1)	66^¶ (6.7) (62, 67.3, 71.1)
Sex; n(%) female	45,971(55.7)	2323(48.8^¶)	3923(61.1^¶)
Ethnicity; n(%) white^‡	79,570(96.4)	4673(98.2^¶)	6307(98.3^¶)
Smoking status^†:
– Never; n(%)	47,774(57.9)	2548(53.5^¶)	3531(55^¶)
– Previous; n(%)	29,225(35.4)	1919(40.3)	2544(39.6)
– Current; n(%)	5392(6.5)	282(5.9)	331(5.2)
– Missing; n(%)	189(0.2)	12(0.3)	13(0.2)
Alcohol intake frequency^†:
– ≤1–3 times a month; n(%)	22,311(27)	1154(24.2^¶)	1718(26.8^¶)
– 1–4 times a week; n(%)	42,370(51.3)	2433(51.1)	3188(49.7)
– Daily or almost daily: n(%)	17,836(21.6)	1171(24.6)	1509(23.5)
– Missing: n(%)	63(0.08)	3(0.06)	4(0.06)
Townsend deprivation index:
– Mean (SD) (25th, 50th, 75th percentile)	-1.7 (2.8) (-3.8, -2.4, -0.1)	-2^¶ (2.7) (-3.9, -2.7, -0.6)	-2^¶ (2.6) (-3.9, -2.7, -0.6)
– Missing; %	99 (0.1)	4 (0.1)	4 (0.1)
BMI group^†:
– Underweight (<18.5); n(%)	507(0.6)	20^¶(0.4)	45(0.7)
– Normal (18.5–<25); n(%)	32,107(38.9)	1777(37.3)	2568(40)
– Overweight (25–<30); n(%)	33,706(40.8)	2044(42.9)	2573(40.1)
– Obese (≥30); n(%)	16,105(19.5)	913(19.2)	1223(19.1)
– Missing; n(%)	155(0.2)	7(0.1)	10(0.2)
Diabetes^§:
– No; n(%)	79,243(96)	4529(95.1^¶)	6136(95.6)
– Yes; n(%)	3202(3.9)	221(4.6)	276(4.3)
– Missing; n(%)	135(0.2)	11(0.2)	7(0.1)
Heart attack^§:
– No; n(%)	81,156(98.3)	4646(97.6^¶)	6288(98)
– Yes; n(%)	1316(1.6)	108(2.3)	121(1.9)
– Missing; n(%)	108(0.1)	7(0.1)	10(0.2)
Angina^§:
– No; n(%)	80,662(97.7)	4605(96.7^¶)	6248(97.3)
– Yes; n(%)	1810(2.2)	149(3.1)	161(2.5)
– Missing; n(%)	108(0.1)	7(0.1)	10(0.2)
Stroke^§:
– No; n(%)	81,606(98.8)	4681(98.3^¶)	6340(98.8)
– Yes; n(%)	866(1)	73(1.5)	69(1.1)
– Missing; n(%)	108(0.1)	7(0.1)	10(0.2)
High blood pressure^§:
– No; n(%)	63,130(76.4)	3338(70.1^¶)	4591(71.5^¶)
– Yes; n(%)	19,342(23.4)	1416(29.7)	1818(28.3)
– Missing; n(%)	108(0.1)	7(0.1)	10(0.2)
Deep vein thrombosis^§:
– No; n(%)	81,138(98.3)	4649(97.6^¶)	6244(97.3^¶)
– Yes; n(%)	1360(1.6)	108(2.3)	169(2.6)
– Missing; n(%)	82(0.1)	4(0.08)	6(0.09)
Blood clot in the lung^§:
– No; n(%)	81,934(99.2)	4720(99.1)	6339(98.8^¶)
– Yes; n(%)	564(0.7)	37(0.8)	74(1.2)
– Missing; n(%)	82(0.1)	4(0.08)	6(0.09)
Emphysema/chronic bronchitis^§:
– No; n(%)	81,571(98.8)	4677(98.2^¶)	6314(98.4^¶)
– Yes; n(%)	927(1.1)	80(1.7)	99(1.5)
– Missing; n(%)	82(0.1)	4(0.08)	6(0.09)
Asthma^§:
– No; n(%)	72,730(88.1)	4245(89.2^¶)	5679(88.5)
– Yes; n(%)	9768(11.8)	512(10.8)	734(11.4)
– Missing; n(%)	82(0.1)	4(0.08)	6(0.09)
Hay fever/allergic rhinitis, or eczema^§:
– No; n(%)	59,866(72.5)	3596(75.5^¶)	4835(75.3^¶)
– Yes; n(%)	22,632(27.4)	1161(24.4)	1578(24.6)
– Missing; n(%)	82(0.1)	4(0.08)	6(0.09)

^† Values at the latest assessment date.

^‡ Other categories include mixed, Asian or Asian British, black or black British, Chinese, and other.

^§ Ever diagnosed by a doctor as reported by the participant at the baseline visit date or the latest reassessment date before the accelerometer data collection.

^¶ Indicates statistically significant difference from the cancer-free cohort at α level 0.05. χ² test was used for categorical variables, and Mann–Whitney U test was used for continuous variables.

MVPA: Moderate-to-vigorous physical activity; SD: Standard deviation.

In univariable analyses, participants in the current cancer group spent, on average, 13 fewer minutes in MVPA per day compared with participants in the cancer-free group (average minutes in MVPA per day: -13.343; 95% CI: -14.846 to -11.84; p < 0.001) (Table 2). The difference in univariable analysis between the cancer survivor and cancer-free groups was slightly smaller (average minutes in MVPA per day: -10.834; 95% CI: -12.14 to -9.527; p < 0.001). In the multivariable linear regression, being in the current cancer versus cancer-free group was associated with fewer minutes in MVPA per day, although the difference was smaller compared with the univariable analysis (average minutes in MVPA per day: -6.57; 95% CI: -7.988, -5.151; p < 0.001). Similarly, after controlling for potential confounders, the cancer survivor group had lower MVPA compared with the cancer-free group, but it was also a smaller difference compared with the univariable analysis (average minutes in MVPA per day: -4.461; 95% CI: -5.697 -3.225; p < 0.001). Variables related to exposure to hazardous materials were excluded from the multivariate model due to high missingness (≥38%). Results from univariate analyses from these variables are included in Supplementary Table 2.

Table 2. Regression results.

Covariates	Continuous/categories	Average minutes in MVPA per day (95% CI)
		Univariable analysis	Multivariable analysis (current cancer vs cancer free; n = 86,423)	Multivariable analysis (cancer survivor vs cancer free; n = 88,063)
Cancer status	Cancer free	Ref.	Ref.	Ref.
	Cancer current	-13.343 (-14.846 to -11.84) p < 0.001^†	-6.57 (-7.988 to -5.151) p < 0.001^†	NA
	Cancer survivor	-10.834 (-12.14 to -9.527) p < 0.001^†	NA	-4.461 (-5.697 to -3.225) p < 0.001^†
Age at accelerometer study end date	Continuous (per year)	-1.714 (-1.754 to -1.673) p < 0.001^†	-1.544 (-1.588 to -1.5) p < 0.001^†	-1.555 (-1.598 to -1.512) p < 0.001^†
Sex	Female	Ref.	Ref.	Ref.
	Male	-5.026 (-5.687 to -4.364) p < 0.001^†	0.074 (-0.597 to 0.745) p = 0.829	-0.008 (-0.673 to 0.657) p = 0.981
Ethnicity	White	Ref.	Ref.	Ref.
	Mixed	9.112 (4.676 to 13.548) p < 0.001^†	1.264 (-3.01 to 5.538) p = 0.562	1.839 (-2.376 to 6.055) p = 0.392
	Asian or Asian British	-5.28 (-8.671 to -1.889) p = 0.002^†	-7.684 (-10.965 to -4.403) p < 0.001^†	-7.787 (-11.056 to -4.518) p < 0.001^†
	Black or black British	7.689 (4.092 to 11.287) p < 0.001^†	5.942 (2.471 to 9.413) p < 0.001^†	5.103 (1.64 to 8.567) p = 0.004^†
	Chinese	24.699 (17.727 to 31.671) p < 0.001^†	14.881 (8.194 to 21.567) p < 0.001^†	14.549 (7.868 to 21.231) p < 0.001^†
	Other	7.937 (3.466 to 12.407) p < 0.001^†	4.583 (0.274 to 8.891) p = 0.037^†	6.361 (2.069 to 10.654) p = 0.004^†
Smoking status	Never	Ref.	Ref.	Ref.
	Previous	-4.054 (-4.752 to -3.355) p < 0.001^†	1.919 (1.217 to 2.621) p < 0.001^†	1.951 (1.256 to 2.645) p < 0.001^†
	Current	-10.733 (-12.101 to -9.365) p < 0.001^†	-11.428 (-12.773 to -10.083) p < 0.001^†	-11.547 (-12.884 to -10.21) p < 0.001^†
Alcohol intake frequency	≤1–3 times a month	Ref.	Ref.	Ref.
	1–4 times a week	7.462 (6.68 to 8.244) p < 0.001^†	4.972 (4.191 to 5.753) p < 0.001^†	4.973 (4.201 to 5.745) p < 0.001^†
	Daily or almost daily	4.504 (3.559 to 5.45) p < 0.001^†	5.117 (4.157 to 6.077) p < 0.001^†	5.182 (4.233 to 6.13) p < 0.001^†
BMI (kg/m^2)	Normal (18.5–<25)	Ref.	Ref.	Ref.
	Underweight (<18.5)	9.305 (5.183 to 13.426) p < 0.001^†	7.724 (3.576 to 11.873) p < 0.001^†	8.995 (4.944 to 13.045) p < 0.001^†
	Overweight (25–<30)	-15.662 (-16.378 to -14.947) p < 0.001^†	-12.613 (-13.346 to -11.88) p < 0.001^†	-12.85 (-13.575 to -12.125) p < 0.001^†
	Obese (≥30)	-33.007 (-33.894 to -32.12) p < 0.001^†	-28.145 (-29.076 to -27.213) p < 0.001^†	-28.339 (-29.26 to -27.418) p < 0.001^†
Townsend index	Continuous	-0.082 (-0.199 to 0.035) p = 0.17	-0.166 (-0.282 to -0.049) p = 0.005^†	-0.182 (-0.298 to -0.066) p = 0.002^†
Diabetes diagnosed by doctor	No	Ref.	Ref.	Ref.
	Yes	-31.022 (-32.7 to -29.344) p < 0.001^†	-14.435 (-16.136 to -12.734) p < 0.001^†	-14.197 (-15.885 to -12.51) p < 0.001^†
Heart attack	No	Ref.	Ref.	Ref.
	Yes	-28.297 (-30.875 to -25.719) p < 0.001^†	-8.324 (-10.997 to -5.651) p < 0.001^†	-8.323 (-10.983 to -5.663) p < 0.001^†
Angina	No	Ref.	Ref.	Ref.
	Yes	-28.308 (-30.514 to -26.103) p < 0.001^†	-7.18 (-9.479 to -4.88) p < 0.001^†	-6.785 (-9.078 to -4.492) p < 0.001^†
Stroke	No	Ref.	Ref.	Ref.
	Yes	-26.711 (-29.897 to -23.526) p < 0.001^†	-12.299 (-15.442 to -9.155) p < 0.001^†	-13.04 (-16.183 to -9.897) p < 0.001^†
High blood pressure	No	Ref.	Ref.	Ref.
	Yes	-18.874 (-19.634 to -18.114) p < 0.001^†	-5.649 (-6.448 to -4.85) p < 0.001^†	-5.569 (-6.36 to -4.778) p < 0.001^†
Deep vein thrombosis	No	Ref.	Ref.	Ref.
	Yes	-11.16 (-13.671 to -8.65) p < 0.001^†	-0.042 (-2.602 to 2.518) p = 0.975	-0.189 (-2.697 to 2.32) p = 0.883
Blood clot in the lung	No	Ref.	Ref.	Ref.
	Yes	-14.892 (-18.782 to -11.003) p < 0.001^†	-6.546 (-10.524 to -2.567) p = 0.001^†	-5.818 (-9.677 to -1.958) p = 0.003^†
Emphysema/chronic bronchitis	No	Ref.	Ref.	Ref.
	Yes	-18.791 (-21.835 to -15.747) p < 0.001^†	-7.559 (-10.583 to -4.535) p < 0.001^†	-8.324 (-11.32 to -5.328) p < 0.001^†
Asthma	No	Ref.	Ref.	Ref.
	Yes	-1.01 (-2.032 to 0.012) p = 0.053	-0.249 (-1.269 to 0.77) p = 0.632	-0.485 (-1.493 to 0.524) p = 0.346
Hay fever/allergic rhinitis or eczema	No	Ref.	Ref.	Ref.
	Yes	1.569 (0.829 to 2.31) p < 0.001^†	-2.151 (-2.889 to -1.413) p < 0.001^†	-2.074 (-2.805 to -1.343) p < 0.001^†

^† Indicates statistically significant result at α level 0.05.

MVPA: Moderate-to-vigorous physical activity; NA: Not applicable.

Changing the cut-off value for years since the latest cancer diagnosis did not substantially change the observed associations, although lowering the cut-off value (i.e., including more recent cases in the current cancer group) tended to increase the difference in PA levels with the cancer-free group in general (Table 3).

Table 3. Sensitivity analysis results.

Years since latest cancer diagnosis	Average daily minutes spent in MVPA (95% CI, p-value) (Reference is cancer-free group in all analyses)
	Cancer current, univariable analysis (n = 87,341)	Cancer current, multivariable analysis^† (n = 86,423)	Cancer survivor, univariable analysis (n = 88,999)	Cancer survivor, multivariable analysis^† (n = 88,063)
4-year cut off	-13.93 (-15.562 to -12.298) p < 0.001	-6.899 (-8.437 to -5.36) p < 0.001	-10.74 (-11.979 to -9.501) p < 0.001	-4.443 (-5.617 to -3.27) p < 0.001
3-year cut off	-15.153 (-16.988 to -13.318) p < 0.001	-7.923 (-9.651 to -6.195) p < 0.001	-10.641 (-11.817 to -9.466) p < 0.001	-4.341 (-5.455 to -3.227) p < 0.001
2-year cut off	-14.748 (-16.958 to -12.538) p < 0.001	-7.711 (-9.786 to -5.637) p < 0.001	-11.215 (-12.328 to -10.103) p < 0.001	-4.75 (-5.806 to -3.693) p < 0.001
1-year cut off	-16.093 (-19.219 to -12.966) p < 0.001	-8.656 (-11.586 to -5.725) p < 0.001	-11.487 (-12.542 to -10.431) p < 0.001	-4.979 (-5.982 to -3.976) p < 0.001

^† After controlling for age at accelerometer study end date, BMI category, smoking status, alcohol intake frequency, ethnicity (white, mixed, Asian or Asian British, black or black British, Chinese, other), Townsend index, and binary variables for diagnosis by a doctor of following co-morbidities: diabetes, heart attack, angina, stroke, high blood pressure, deep vein thrombosis, blood clot in the lung, emphysema/chronic bronchitis, asthma, hay fever/allergic rhinitis or eczema.

MVPA: Moderate-to-vigorous physical activity.

There were 183 deaths among the 4712 patients in the current cancer group during the follow-up period. In the survival analysis, the hazard ratio of all-cause mortality associated with an additional hour of MVPA per day was 0.518 (95% CI: 0.404–0.663; p < 0.001) or 0.94 (95% CI: 0.91–0.96; p < 0.001) for a 6-min increase in time in MVPA per day (Model 1 in Table 4). When hospitalization within the past 3 months of accelerometer data collection was included in the regression model (Model 2 in Table 4), the hazard ratio was attenuated to 0.562 (95% CI: 0.44–0.719; p < 0.001), which corresponds to a hazard ratio of 0.94 (95% CI: 0.92–0.98; p < 0.001) for a 6-min increase in time in MVPA – this was still statistically significant.

Table 4. Survival analysis results – all-cause death among the current cancer group.

Covariates	Continuous/categories	Number of participants, number of events	Cox regression: Model 1 (current cancer patients, n = 4761)	Cox regression: Model 2 multivariable analysis (current cancer patients, n = 4761)
Average hours of MVPA per day	Continuous (per day)	N: 4712, n: 183	0.518 (0.404, 0.663) p < 0.001^†	0.562 (0.44, 0.719) p < 0.001^†
Hospitalized within the past 3 months	No	N: 4150, n: 131	–	Ref.
	Yes	N: 562, n: 52	–	2.295 (1.646, 3.2) p < 0.001^†
Age at accelerometer study end date	Continuous (per year)	N: 4712, n: 183	1.021 (0.994, 1.048) p < 0.13	1.023 (0.996, 1.051) p < 0.093
Sex	Female	N: 2305, n: 70	Ref.	Ref.
	Male	N: 2407, n: 113	1.324 (0.959, 1.827) p < 0.088	1.315 (0.952, 1.819) p < 0.097
Ethnicity	White	N: 4644, n: 178	Ref.	Ref.
	Mixed	N: 13, n: 2	3.103 (0.736, 13.085) p < 0.123	3.778 (0.893, 15.982) p < 0.071
	Asian or Asian British	N: 19, n: 0	N/A (N/A, N/A) p < N/A	N/A (N/A, N/A) p < N/A
	Black or black British	N: 15, n: 2	2.647 (0.621, 11.271) p < 0.188	1.945 (0.443, 8.533) p < 0.378
	Chinese	N: 6, n: 1	6.73 (0.804, 56.309) p < 0.079	7.32 (0.866, 61.892) p < 0.068
	Other	N: 15, n: 0	N/A (N/A, N/A) p < N/A	N/A (N/A, N/A) p < N/A
Smoking status	Never	N: 2533, n: 73	Ref.	Ref.
	Previous	N: 1900, n: 86	1.372 (0.993, 1.897) p < 0.056	1.308 (0.945, 1.81) p < 0.105
	Current	N: 279, n: 24	2.205 (1.347, 3.61) p < 0.002^†	2.151 (1.312, 3.528) p < 0.002^†
Alcohol intake frequency	≤1–3-times a month	N: 1137, n: 50	Ref.	Ref.
	1–4-times a week	N: 2416, n: 86	0.922 (0.637, 1.336) p < 0.669	0.938 (0.647, 1.362) p < 0.738
	Daily or almost daily	N: 1159, n: 47	0.862 (0.562, 1.322) p < 0.496	0.879 (0.572, 1.351) p < 0.556
BMI (kg/m^2)	Normal (18.5–<25)	N: 1767, n: 53	Ref.	Ref.
	Underweight (<18.5)	N: 20, n: 1	1.281 (0.156, 10.52) p < 0.818	1.476 (0.178, 12.258) p < 0.719
	Overweight (25–<30)	N: 2021, n: 81	1.043 (0.73, 1.489) p < 0.817	1.036 (0.724, 1.481) p < 0.848
	Obese (≥30)	N: 904, n: 48	1.039 (0.679, 1.591) p < 0.859	1.021 (0.666, 1.566) p < 0.924
Townsend index	Continuous	N: 4712, n: 183	1.072 (1.02, 1.128) p < 0.007^†	1.073 (1.02, 1.128) p < 0.006^†
Diabetes diagnosed by doctor	No	N: 4492, n: 165	Ref.	Ref.
	Yes	N: 220, n: 18	1.221 (0.728, 2.047) p < 0.449	1.275 (0.76, 2.138) p < 0.357
Heart attack	No	N: 4605, n: 175		Ref.
	Yes	N: 107, n: 8	1.137 (0.534, 2.424) p < 0.739	1.117 (0.526, 2.373) p < 0.773
Angina	No	N: 4564, n: 172	Ref.	Ref.
	Yes	N: 148, n: 11	1.214 (0.635, 2.322) p < 0.558	1.135 (0.595, 2.166) p < 0.701
Stroke	No	N: 4643, n: 180		Ref.
	Yes	N: 69, n: 3	0.777 (0.247, 2.443) p < 0.666	0.775 (0.246, 2.44) p < 0.663
High blood pressure	No	N: 3314, n: 104	Ref.	Ref.
	Yes	N: 1398, n: 79	1.318 (0.96, 1.809) p < 0.087	1.279 (0.93, 1.757) p < 0.13
Deep vein thrombosis	No	N: 4605, n: 180	Ref.	Ref.
	Yes	N: 107, n: 3	0.799 (0.254, 2.52) p < 0.702	0.842 (0.267, 2.651) p < 0.769
Blood clot in the lung	No	N: 4676, n: 183	Ref.	Ref.
	Yes	N: 36, n: 0	N/A (N/A, N/A) p < N/A	N/A (N/A, N/A) p < N/A
Emphysema/chronic bronchitis	No	N: 4632, n: 179	Ref.	Ref.
	Yes	N: 80, n: 4	0.823 (0.3, 2.255) p < 0.704	0.84 (0.306, 2.305) p < 0.734
Asthma	No	N: 4207, n: 163	Ref.	Ref.
	Yes	N: 505, n: 20	1.028 (0.636, 1.663) p < 0.909	1.039 (0.642, 1.679) p < 0.877
Hay fever/allergic rhinitis or eczema	No	N: 3559, n: 140	Ref.	Ref.
	Yes	N: 1153, n: 43	1 (0.7, 1.427) p < 0.998	0.964 (0.675, 1.376) p < 0.84

^† Indicates statistically significant result at α level 0.05.

MVPA: Moderate-to-vigorous physical activity; N: Number of participants; n: Number of events; N/A: Not applicable.

Discussion

Our analysis suggests that patients with a history of cancer have lower levels of PA compared with patients without cancer, and that the difference is greater for patients whose diagnosis is within the last 5 years. Furthermore, lower levels of PA associated with a prior cancer diagnosis are associated with increased risk of mortality.

The recommendation of 150 min of MVPA per week reflects the time in MVPA above each individual’s baseline PA levels [27], such that the recommended total time in MVPA as measured by accelerometery corresponds to approximately 1000 min per week [28]. Time in MVPA per week in the cancer-free cohort was 71% of this level, and 21.4% of participants in this cohort achieved >1000 min of MVPA per week. The mean MVPA for the cancer survivor cohort accounted for 63% of the 1000 min, and 16.4% of participants recorded time in MVPA above this level. The mean MVPA for the current cancer cohort was 61% of the 1000 min, and 14.7% of participants had time in MVPA above the threshold. These results highlight inadequate levels of PA among cancer survivors and emphasize that patients should engage in PA and to reap the benefits.

The absolute MVPA levels and differences observed between cancer survivors and the cancer-free group in this study were greater compared with data reported in previous work. Thraen-Borowski et al. [10] noted that cancer survivors engaged in only two fewer minutes of MVPA per day compared with the cancer-free group [10]. The time spent in MVPA by cancer survivors and the cancer-free group reported was only 16 and 18 min per day, respectively [10]. In our study, the current cancer group averaged 88 min of MVPA per day, the cancer survivor group averaged 90 min and the cancer-free group averaged 101 min. The observed difference might be due to the accelerometer brand (Actigraph AM-7164) used by Thraen-Borowski et al. and their definition of MVPA, which relied on ‘counts-per-minute’ as defined by Actigraph’s algorithm [10].

After adjusting for potential confounding factors, including BMI, smoking and history of comorbidities, the difference in time in MVPA per week between the current cancer and cancer-free groups dropped by half to 42 min per week (6 min per day), which was similar to a recent study that also used UK Biobank data. Barker et al. [9] reported geometric average weekly MVPA of 705 min for their ‘no chronic disease’ cohort compared with 640 min for ‘all malignant cancer but without any other chronic disease group’, which indicates a difference of 65 min per week. Our study extends this previous work to demonstrate that the difference is potentially greater for patients with a relatively recent diagnosis of cancer compared with longer term survivors.

Furthermore, our analysis indicates that the adjusted difference in PA between those with and without a prior cancer diagnosis was associated with lower survival among patients with current cancer, after adjusting for potential confounders. These results are in line with previous studies demonstrating reduced mortality risk among more physically active cancer survivors [29–31], and importantly suggest small changes in daily activity may have an impact on outcomes for patients with cancer. These findings further highlight the potential value to develop interventions that adequately target and support activity levels among cancer survivors.

Limitations

Our study has several important limitations. PA levels were measured for only 1 week and were assumed as representative of participants’ overall PA levels. The UK Biobank does not include information on cancer treatments; therefore, we used time since latest diagnosis as a proxy for whether a participant might be undergoing treatment. We also did not have information on disease severity or stage for the participants in the cancer groups, which might explain some of the variations in PA outcomes. Although we included the Townsend index score as a covariate in our analyses, we did not include education level, which may have a correlation with fitness. This study used a cross-sectional design; therefore, the results cannot be interpreted as evidence of causality, even though they do indicate that UK Biobank participants with a history of cancer tend to be less active. UK Biobank participants are not representative of the broader UK population with respect to some important demographic and clinical characteristics. They are less likely to be obese, current smokers and drink alcohol daily compared with the general UK population [32]. Furthermore, the all-cause mortality and total cancer incidence are lower (46.2 and 11.8% lower in men, respectively, and 55.5 and 18.1% lower in women, respectively) compared with the general UK population, suggesting there is selection bias [32]. Accelerometer study participants (with or without a history of cancer) represent a subset of the UK Biobank participants who were willing and able to join a study on objective measurement of PA, which reduces the generalizability of our results to the cancer survivor population, because healthier cancer survivors may be more likely to have joined the study, as well as the general population. Although this study used the latest available data, only a small minority of UK Biobank participants joined reassessment visits. If the values for variables, such as BMI, smoking status or presence of comorbidities, changed at the time of accelerometer study, there may be a regression dilution bias, which is caused by measurement error in covariates that tends to decrease their association with the outcome.

Conclusion

Patients with a cancer diagnosis in the last 5 years and longer term cancer survivors had lower PA levels compared with the population without any cancer diagnosis. The lower levels of PA associated with a cancer history might be associated with increased mortality among more recently diagnosed patients. Additional analyses are needed to understand the effect of cancer stage and treatment on PA. In addition, the impact on hospitalization, quality of life and mortality of lower levels of PA among current cancer patients and cancer survivors should be further investigated, and ways to improve PA in cancer patients must be explored.

Executive summary

• We used data from the UK Biobank, a large prospective cohort study, to identify the differences in physical activity (PA) levels between participants with and without cancer, and to estimate the all-cause mortality hazard associated with PA among participants more recently diagnosed with cancer.

• We created three cohorts based on cancer status: participants with a cancer diagnosis within the last 5 years (current cancer), before 5 years (cancer survivor) and without any cancer record (cancer free).

• We extracted daily average minutes in moderate-to-vigorous PA (MVPA) from accelerometer data and compared it for the cancer groups versus the cancer-free group using multivariable regression.

• We used a Cox regression to estimate the relationship between all-cause mortality and MVPA.

• Average time spent daily in MVPA was 88 min for the current cancer group (N = 4761), 90 min for the cancer survivor group (N = 6352) and 101 min for the cancer-free group (N = 82,580).

• Compared with the cancer-free group, the current cancer and cancer survivor groups had fewer minutes in MVPA per day in adjusted analyses (6 and 4 min fewer, respectively).

• All-cause mortality was reduced by 6% for a 6-min increase in time with MVPA.

• Patients with a history of cancer were less active compared with the cancer-free group.

• The lower levels of MVPA were associated with increased mortality among more recently diagnosed patients.

• Our results suggest even small increases in daily activity can help reduce mortality risk for patients with cancer, supporting the need to develop interventions that adequately target and support activity levels among cancer survivors.

Ethical conduct of research

The authors state that they have obtained appropriate institutional review board approval or have followed the principles outlined in the Declaration of Helsinki for all human or animal experimental investigations. In addition, for investigations involving human subjects, informed consent has been obtained from the participants involved.

Financial&competing interests disclosure

S Ramagopalan, L McDonald, R Carroll, P Thakkar, B Malcolm and F Mehmud are employees of Bristol-Myers Squibb. M Oguz, N Dhalwani and E Merinopoulou are employees of Evidera, and F Yang and A Cox were employees of Evidera at the time of conduct of this study. Evidera received funding from Bristol-Myers Squibb for this study. This research has been conducted using the UK Biobank Resource under application number 44513. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.

Additional information

Funding

S Ramagopalan, L McDonald, R Carroll, P Thakkar, B Malcolm and F Mehmud are employees of Bristol-Myers Squibb. M Oguz, N Dhalwani and E Merinopoulou are employees of Evidera, and F Yang and A Cox were employees of Evidera at the time of conduct of this study. Evidera received funding from Bristol-Myers Squibb for this study. This research has been conducted using the UK Biobank Resource under application number 44513. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed. No writing assistance was utilized in the production of this manuscript.