SoylentNews

New Study Reveals 'Startling' Risk of Stroke

posted by Fnord666 on Thursday December 20 2018, @04:07PM   
from the in-your-hands dept.

upstart writes:

Submitted via IRC for Bytram

New study reveals 'startling' risk of stroke

Globally, one in four people over age 25 is at risk for stroke during their lifetime, according to a new scientific study.

Researchers found a nearly five-fold difference in lifetime stroke risk worldwide, with the highest risk in East Asia and Central and Eastern Europe, and lowest in sub-Saharan Africa. The lifetime stroke risk for 25-year-olds in 2016 ranged from 8% to 39%, depending on where they live; people in China have the highest risk.

"Our findings are startling," said Dr. Gregory Roth, Assistant Professor of Health Metrics Sciences at the Institute for Health Metrics and Evaluation (IHME) at the University of Washington, and senior author on the study. "It is imperative that physicians warn their patients about preventing strokes and other vascular diseases at earlier points in patients' lives. We found extremely high lifetime risk for stroke, and based on other research we evaluated, it is clear that younger adults need to think about long-term health risks. They can make a real difference by eating healthier diets, exercising regularly, and avoiding tobacco and alcohol." 

The study, "Global, Regional, and Country-Specific Lifetime Risks of Stroke, 1990-2016," was published today in The New England Journal of Medicine.

[...] The burden of stroke among adults is largely dependent on modifiable risk factors and the characteristics of health systems. Therefore, the study's findings may be useful for long-term planning, especially in terms of prevention and public education.

[...] "This important paper provides reliable data on current lifetime risks across the world for different types of stroke, as well as providing countries with valuable insights into the burden of stroke," said Dr. Peter Rothwell, Head of the Centre for the Prevention of Stroke and Dementia and Professor of Clinical Neurology at the University of Oxford. "These data and insights can be used to prioritize and target strategies for prevention. I hope this important work will be continued so that these trends can be mapped in future decades."

---

Original Submission

• Try this Try this (Score: 1, Interesting) by Anonymous Coward on Thursday December 20 2018, @05:24PM (10 children)

  by Anonymous Coward on Thursday December 20 2018, @05:24PM (#776855)

  Open up R (comes with most linux distros), and paste:
   

  set.seed(12345)
    treatment = c(rep(1, 4), rep(0, 4))
    gender1   = rep(c(1, 0), 4)
    gender2   = rep(c(0, 1), 4)
    result    = rnorm(8)

    summary(lm(result ~ treatment*gender1))
    summary(lm(result ~ treatment*gender2))

  Here are the results:

  >   summary(lm(result ~ treatment*gender1))

  Call:
  lm(formula = result ~ treatment * gender1)

  Residuals:
         1        2        3        4        5        6        7        8
    0.34742  0.58148 -0.34742 -0.58148 -0.01211 -0.77089  0.01211  0.77089

  Coefficients:
                    Estimate Std. Error t value Pr(>|t|)
  (Intercept)        -1.0471     0.5131  -2.041   0.1109
  treatment           1.1751     0.7257   1.619   0.1807
  gender1             1.6651     0.7257   2.294   0.0834 .
  treatment:gender1  -1.5549     1.0263  -1.515   0.2043
  ---
  Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

  Residual standard error: 0.7257 on 4 degrees of freedom
  Multiple R-squared:  0.5955,    Adjusted R-squared:  0.2921
  F-statistic: 1.963 on 3 and 4 DF,  p-value: 0.2617

  >   summary(lm(result ~ treatment*gender2))

  Call:
  lm(formula = result ~ treatment * gender2)

  Residuals:
         1        2        3        4        5        6        7        8
    0.34742  0.58148 -0.34742 -0.58148 -0.01211 -0.77089  0.01211  0.77089

  Coefficients:
                    Estimate Std. Error t value Pr(>|t|)
  (Intercept)         0.6180     0.5131   1.204   0.2948
  treatment          -0.3799     0.7257  -0.523   0.6283
  gender2            -1.6651     0.7257  -2.294   0.0834 .
  treatment:gender2   1.5549     1.0263   1.515   0.2043
  ---
  Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

  Residual standard error: 0.7257 on 4 degrees of freedom
  Multiple R-squared:  0.5955,    Adjusted R-squared:  0.2921
  F-statistic: 1.963 on 3 and 4 DF,  p-value: 0.2617

  You can see just by coding male/female as 1/0 instead of 0/1, the effect of the treatment goes from positive (1.1751) to negative (-0.3799). This is the type of stuff they are doing to generate these "startling" results.

  Starting Score:	0		points
  Moderation		+1
  Interesting=1, Total=1
  Extra 'Interesting' Modifier		0
  Total Score:		1

• Re:Try this Re:Try this (Score: 0) by Anonymous Coward on Thursday December 20 2018, @06:35PM (7 children)

  by Anonymous Coward on Thursday December 20 2018, @06:35PM (#776895)

  Haha, a new Nature program: When Coders Try to Science!

  Parent

  • Re:Try this Re:Try this (Score: 0) by Anonymous Coward on Thursday December 20 2018, @06:53PM (6 children)

    by Anonymous Coward on Thursday December 20 2018, @06:53PM (#776906)

    This ain't science. It's NHST.

    If it were science you would see people checking predictions made earlier against data they just collected. You would also see there are plans to attempt to exactly replicate this study by some other group and make sure the same results are returned.

    Parent

    • Re:Try this Re:Try this (Score: 0) by Anonymous Coward on Thursday December 20 2018, @07:41PM (5 children)

      by Anonymous Coward on Thursday December 20 2018, @07:41PM (#776942)

      It's NHST

      The authors did not perform any statistical test.

      Parent

      • Re:Try this Re:Try this (Score: 0) by Anonymous Coward on Thursday December 20 2018, @07:46PM (4 children)

        by Anonymous Coward on Thursday December 20 2018, @07:46PM (#776943)

        Point estimates and 95% uncertainty intervals representing the 2.5th and 97.5th percentiles around the estimate were used to compare results between groups. Differences in the estimates of the risk of stroke were considered to be significant when the 95% uncertainty intervals did not overlap or when the 95% uncertainty interval for relative percentage change did not include zero.

        [...]

        The estimates of global lifetime risk of stroke increased from 22.8% in 1990 to 24.9% in 2016, representing a relative increase of 8.9% (95% uncertainty interval, 6.2 to 11.5) (Table 1, and Table S4 in the Supplementary Appendix); the difference is significant, as reflected by the exclusion of zero in the uncertainty interval.

        https://www.nejm.org/doi/10.1056/NEJMoa1804492 [nejm.org]

        But yea, this paper doesn't really do the whole "adjust/correct for confounders" linear/logistic regression thing.

        Parent

        • Re:Try this Re:Try this (Score: 0) by Anonymous Coward on Thursday December 20 2018, @08:24PM (3 children)

          by Anonymous Coward on Thursday December 20 2018, @08:24PM (#776956)

          They did not perform a test.
          They did model the data:

          An ensemble model was used to estimate a continuous time-series for mortality by age, sex, country
          (developed or developing), and year. Country-level covariates associated with stroke were used
          in the model and out-of-sample validity testing was used to assess model performance. 95%
          uncertainty intervals (UI) were estimated using 1000 draws from the posterior distribution for
          each age-sex-country group. Disease prevalence was estimated using DisMod state-transition
          disease modeling software[9] and Bayesian staistical models. IS and HS were modelled
          separately and combined.

          Parent

          • Re:Try this Re:Try this (Score: 0) by Anonymous Coward on Friday December 21 2018, @01:20AM (2 children)

            by Anonymous Coward on Friday December 21 2018, @01:20AM (#777038)

            There you go, same thing. Their coefficients will change if they add more covariates, or add/remove some interactions, or decide to log transform a feature, or any of a million arbitrary choices.

            Parent

            • Re:Try this Re:Try this (Score: 0) by Anonymous Coward on Friday December 21 2018, @04:15AM (1 child)

              by Anonymous Coward on Friday December 21 2018, @04:15AM (#777090)

              Just admit you were wrong about the testing and lack of modeling data.
              You can argue about their methods afterwards, but there is no need to lie.

              Parent

              • Re:Try this (Score: 0) by Anonymous Coward on Friday December 21 2018, @02:59PM

                by Anonymous Coward on Friday December 21 2018, @02:59PM (#777208)

                I wasn't wrong, I quoted their NHST and put it in bold for you. Your quote shows they also did the "adjust for confounders" modelling the R script is an example of.

                Parent

• I just see a bunch line noise I just see a bunch line noise (Score: 0) by Anonymous Coward on Thursday December 20 2018, @07:13PM (1 child)

  by Anonymous Coward on Thursday December 20 2018, @07:13PM (#776918)

  You people use this thing to do real work?

  Parent

  • Re:I just see a bunch line noise (Score: 0) by Anonymous Coward on Thursday December 20 2018, @07:37PM

    by Anonymous Coward on Thursday December 20 2018, @07:37PM (#776941)

    Sure, try running this first

    # Library that colorizes R output
    # https://github.com/jalvesaq/colorout
    git clone https://github.com/jalvesaq/colorout.git
    R CMD INSTALL colorout

    # Run R
    R --no-save

    # R options (eg, dont wrap text prematurely)
    # Run every time or add to Rprofile.site file
    options(width = 800)
    options(help_type = "html")
    require(colorout)

    No one programs in the terminal though, you use sublime text or Rstudio (an R IDE).

    Parent