Statistics and pitfalls of trend analysis in cancer research: a review focused on statistical packages

Abstract

Trend analysis is the analysis using statistical models to estimate and predict potential trends over time, space or any independent continuous-variable. It has been widely used in epidemiology and public health, but much less so in clinical oncology and basic cancer research. Methodological imitations of the chosen statistical package also appear to result in biased or less rigorous interrogation of cancer-related data. We thus review the basic statistics of trend analysis, commonly used commands of statistical packages and the common pitfalls of conducting trend analysis. Four free and 3 commercial statistical-packages were discussed in depth, including Joinpoint, Epi info, R package, Python, SAS, Stata and SPSS. We hope that this review could serve as a practical yet concise guide for using statistical packages for trend analysis in translational and clinical oncology, and help improve the scientific rigor of trend analyses in these fields. The guide, however, may also be applied to other research fields.

Associations of Stay-at-Home Order and Face-Masking Recommendation with Trends in Daily New Cases and Deaths of Laboratory-Confirmed COVID-19 in the United States

“…Further, two piecewise log-linear regression methods were used to rigorously examine trends changes, according to the guidelines on trend analyses of population data and other methodological considerations. [26][27][28][29] Finally, the simulation analysis provides comprehensive estimates of trends changes linked to early-implementation of SAHO at various time points and early-lift of SAHO with various extents. Indeed, the recent upward trends in the USA COVID-19 epidemics are consistent with our simulated scenarios of early SAHO lift (even in some states or counties).…”

“…The District of Columbia was included. *The p-values of comparing the state groups with New York (Wilcoxon rank sum and Student's t test) were all smaller than 0.001. models recommended by the guidelines of the USA's National Center for Health Statistics and other methodological considerations [26][27][28][29]. The simulation analysis demonstrates a much smaller scale of COVID-19 in the USA if SAHO would have been implemented earlier, and a concerning reverse of stable downward trends in COVID-19 daily new cases and deaths if being lifted soon.…”

Screening of differentially expressed genes and identification of AMACR as a prognostic marker in prostate cancer

“…Prostate cancer, the second most common cancer found in male over the world, was estimated to have 191,930 new cases and 33,330 deaths in 2020 in the United States, accounting for 10.6% of all new cancer cases and 5.5% of all cancer deaths respectively (Reynolds, 1992; Xu et al., 2020). Based on the data from 2013 to 2017, 109.8 new cases were diagnosed per 100,000 men with 19.1 deaths.…”

“…Based on the data from 2013 to 2017, 109.8 new cases were diagnosed per 100,000 men with 19.1 deaths. Prostate cancer is very common in male, about 12.1% of men will acquire this cancer in their lifetime, and a higher risk was reported in older men and African American men (Reynolds, 1992; Xu et al., 2020). Patients with prostate cancer have an overall 5‐year relative survival rate of 97.8%.…”

“…Patients with prostate cancer have an overall 5‐year relative survival rate of 97.8%. Patients diagnosed in localised and regional stages have a 5‐year relative survival rate of nearly 100%, while for those in distant stages, this value drops to 30.2% (Reynolds, 1992; Xu et al., 2020). Although prostate cancer grows slowly, an effective early detection is still important for the reduction of mortality.…”