This  presentation  is  about cohort  studies  and  their  role  in  epidemiology. Thanks  to  Nanker  Halbert  smart  and Sarah  wild  for  sharing  their  materials. What  will  this  presentation  cover? We'll  look  at  what  a  cohort  study  is, what  are  its  key  features. We'll  look  at  some  examples  of influential  cohort  studies  which  have improved  our  understanding  of  health  and  disease. We'll  look  at  how  data  are  analysed  in  a  cohort  study. We  look  at  the  strengths  and limitations  of  cohort  studies. And  I'll  highlight  what  to  look  out  for  when you're  critically  appraising  a  cohort  study. This  is  a  recap  from the  previous  presentation  on  cross-sectional  studies. In  that  presentation,  we  highlighted a  key  weakness  of  cross-sectional  studies. Namely  that  because  they  just  collect  data  on exposures  and  outcomes  at  a  single  point  in  time. They  cannot  determine  which  comes  first. Did  the  exposure  preceded the  outcome  or  did  the  outcome  precede  the  exposure? We  use  the  example  of a  cross-sectional  survey  of  London  Transport  Workers, drivers  with  sedentary  jobs, and  bus  conductors  with  more  physically  active  jobs. If  we  had  conducted a  cross-sectional  survey  at  point  B  on  the  slide, we  defined  that  whereas  none  of the  more  active  bus  conductors  had  angina, 9.5  per  cent  of  this  sedentary  bus  drivers  had  angina. This  might  lead  us  to conclude  that  having  a  sedentary  jobs such  as  being  a  bus  conductor increases  the  risk  of  angina. But  we  have  to  be  very  careful  here. We  have  to  remember  that  a  survey  conducted  at point  B  is  just  a  snapshot  in  time. We  don't  know  whether the  bus  drivers  got  angina  because  of their  sedentary  jobs  or  whether  they  chose to  do  sedentary  jobs  because  of  their  poor  health. We  don't  know  which  came  first, the  sedentary  work  or  the  angina, the  exposure  or  the  outcome. This  is  a  key  drawback  for cross-sectional  studies  and  there's  no  way  round  it. If  we  want  to  figure  out  which  came  first,  the  exposure, or  the  outcome,  we  need  a  different  study  design. We  need  to  start  our  study  at  point  a  with a  healthy  cohort  of  workers,  nobody  has  angina. We  then  need  to  follow  up this  cohort  of  workers  over  time, keeping  track  of  their  exposures, what  job  they  do,  whether  they  change  jobs  and  so  on. And  also  keeping  track  of  their  outcomes, whether  they  develop  angina. It's  only  this  sort  of  study,  a  cohort  study, which  enables  us  to  determine  which  comes  first, the  exposure  or  the  outcome. In  other  words,  it's  only  by  using a  cohort  study  that  we  can  establish a  temporal  relationship  between an  exposure  and  an  outcome. The  easiest  way  to  understand the  mechanics  of  a  cohort  study is  by  sketching  it  out  in a  diagram  like  the  one  on  the  slide. At  the  beginning  of  the  study, the  box  on  the  left. All  the  study  participants  should  be free  of  the  outcome  of  interest. Some  of  them  should  have the  exposure  of  interest  and  some  of  them  should not  have  the  exposure  of  interest.  So  e.g. if  we  were  doing  a  cohort  study  to  investigate the  association  between  smoking  and  lung  cancer. At  the  beginning  of  the  study, none  of  the  participants  should  have  lung  cancer. Sum  should  be  smokers  and  some  of them  should  be  nonsmokers. Next,  the  researchers  follow up  the  participants  over  time. At  the  end  of  the  study, the  box  on  the  right, they  count  up  the  number  of  people  with  the  outcome  in the  exposed  group  and  in  the  unexposed  group. So  they  would  count  up the  number  of  smokers  with  and  without lung  cancer  and  the  number  of non-smokers  with  and  without  lung  cancer. If  the  proportion  of  smokers  with  lung  cancer is  bigger  than  the  proportion  of non-smokers  with  lung  cancer. This  would  suggest  that  smoking is  associated  with,  with  lung  cancer. This  type  of  cohort  study  is  called a  prospective  cohort  study. If  something  is  prospective, it  means  it's  expected  to  happen  in  the  future. The  researchers  have  to  wait  for  the  outcome  to  occur. One  of  the  challenges  of  doing  this  sort  of cohort  study  is  that  it  takes  a  long  time, sometimes  decades  to  get  the  results. And  so  it's  very  expensive. There  is  another  alternative. This  is  called  a  retrospective  cohort  study. Instead  of  waiting  for  decades  for  the  outcomes  to  occur, the  researchers  look  back  in  time. To  collect  data  on  the  exposures. Nowadays,  with  the  availability of  routinely  collected  data, this  is  much  easier  than  it  used  to  be. And  the  approach  is  becoming  much more  common  than  it  used  to  be. Here's  an  example  of  how a  retrospective  cohort  study  would  work  in  practise. Let's  say  that  the  researchers  are interested  in  smoking  and  lung  cancer. Again,  they  might  choose  a  group  of  people  who  are  alike in  many  ways  but  differ  with  respect  to  the  exposure. So  e.g.  they  might  choose  as  their  study  population, female  nurses, some  of  whom  smoke  and  some  of  whom  don't  smoke. They  would  then  collect  data  on  the  exposure which  is  smoking  and  the  outcome, which  is  lung  cancer  from the  medical  records  of  this  study  population. It  still  follows  the  same  principles. All  the  participants  are  free  of the  outcome  at  the  start  of  the  study  period. The  difference  is  that  the  study  period begins  sometime  in  the  past. Here's  a  summary  of  the  pros and  cons  of  these  two  approaches. In  general,  prospective  cohort studies  are  more  likely  to have  complete,  unreliable  data. Retrospective  cohort  studies  rely on  routinely  collected  data. So  data  on  important  confounding  factors such  as  smoking  status  are  often  missing. On  the  other  hand,  retrospective  cohort  studies are  quicker,  cheaper, and  easier  to  conduct  and  are  better  for  diseases  with long  latency  periods  because  you  don't  have  to  wait  for years  for  participants  to  develop  the  outcome. Let's  look  at  some  examples  of some  influential  cohort  studies  that  have contributed  to  advances  in our  understanding  about  disease  risk  factors. The  figure  on  this  slide  is  taken  from a  very  famous  cohort  study  on  smoking  and  lung  cancer. Epidemiologists  Richard  Doll  and Austin  Bradford  Hill  used British  doctors  as  their  study  population. This  was  because  they  thought  that  doctors  would  be a  relatively  simple  group  to  keep in  touch  with  and  to  follow  up. In  1951,  they  sent  out  questionnaires  to British  doctors  and  recruited around  40,000  of  them  to  the  study. Numbers  of  female  responders  were  quite  small, so  they  decided  to  restrict  the  study  to  males. They  followed  up  this  group  of  doctors  several times  over  the  next  few  decades. They  published  a  paper  in 1956  showing  that  the  death  rate  from lung  cancer  among  heavy  smokers  was 20  times  the  death  rate  in  non-smokers. They  also  showed  that lung  cancer  death  rates  were  substantially higher  among  cigarette  smokers compared  with  pipe  or  cigar  smokers. They  showed  that  the  heavier  the  smoker, the  more  likely  he  was  to  die  of  lung  cancer. Important  advantage  of  cohort  studies  is that  it's  possible  to  collect  data on  more  than  one  cause  of  death. This  study  was  also  able  to  show  that  men  who  smoked were  more  likely  than  non-smokers to  die  from  coronary  heart  disease. The  last  paper  from  this  study was  published  by  doll  in  2004, just  a  year  before  he  died. By  this  time,  the  idea  that  smoking causes  lung  cancer  was  well  established. However,  the  British  doctors  study  was  still making  a  valuable  contribution  to  our  understanding. The  2004  paper  showed  that  smoking  Lord, life  expectancy  by  an  average  of  ten  years. And  that  half  of  those  who smoked  were  killed  by  their  habit, as  shown  in  the  image  on  the  slide. Another  highly  influential  cohort  study is  the  Framingham  Heart  Study, which  is  still  going  on  more  than  half  a  century  after  it was  set  up  in  the  middle  of  the  20th  century, little  was  known  about the  general  causes  of  heart  disease  and  stroke. But  the  death  rates  for  cardiovascular  disease  had  been increasing  steadily  since  the  beginning of  the  century  in  the  USA. In  common  with  other  developed  countries. In  1950,  the  Framingham  Heart  Study was  set  up  in  the  small  town  of  Framingham, massachusetts  to  identify  the  common  factors that  contribute  to  cardiovascular  disease  by  following its  development  over  a  long  period of  time  in  a  large  group  of participants  who  had  not  yet developed  overt  symptoms  of  cardiovascular  disease. Cardiovascular  disease. Or  suffered  a  heart  attack  or  stroke. The  original  study  involved  over 5,000  men  and  women  between  the  ages  of 30.62  who  have  continued  to  return  to this  study  every  two  years for  a  detailed  medical  history, physical  examination,  and  laboratory  tests. Over  the  subsequent  decades, several  more  cohorts  were  recruited, comprising  the  adult  children and  grandchildren  of  the  original  cohort. And  because  the  original  cohort  was  overwhelmingly  white, and  additional  cohort  was  established  to reflect  the  ethnic  diversity  of  the  population. Over  the  years,  the  Framingham  Heart  Study  has  identified the  major  cardiovascular  risk  factors including  high  blood  pressure, high  blood  cholesterol,  smoking, obesity,  diabetes,  and  physical  inactivity. As  well  as  providing  a  great  deal  of  valuable  information on  the  effects  of related  factors  such  as  blood  triglyceride, an  HDL  cholesterol  levels, age,  gender,  and  psychosocial  issues. This  has  led  to  the  development  of effective  treatment  of  preventative  strategies. The  image  on  the  slide  shows  some  data  from  this  study, which  illustrates  that  although  all  cause cardiovascular  disease  mortality  rates have  fallen  over  the  period  of  the  study, mortality  rates  are  consistently  higher  for those  with  diabetes  compared  to  those  without. The  graph  on  this  slide  is taken  from  a  famous  series  of  studies conducted  in  the  UK  called the  Whitehall  Study  of  British  civil  servants. This  study  was  very  influential in  demonstrating  the  importance  of social  factors  in  determining our  health  and  life  expectancy. The  Whitehall  Study  of civil  servants  was  a  long  term  cohort  study of  the  health  of  men employed  a  civil  servants  in  Britain. The  study  sample  encompassed professional  civil  servants  from  the  top  of the  civil  service  down  to clerical  and  unskilled  staff  at  the  bottom. In  other  words,  there  was  a  social  hierarchy  of participants  with  the  administrators at  the  top  in  pale  purple  on  the  graph, followed  by  the  professional and  executive  staff  in  dark  green, then  the  clerical  staff  in  pale  green, and  ancillary  staff,  drivers, caterers,  cleaners,  and  so  on  in  blue, at  the  bottom  of  the  social  hierarchy, the  graph  shows  the  relative  deaf  death  rates  of these  different  cadres  of civil  servants  at  different  ages. The  first  set  of  bars  shows the  death  rates  among  40  to  64  year  olds. The  next  set  of  bars  shows  the  death  rates  among 65  to  69  year  olds  and  so  on. What  this  study  showed  was  that  not  all  needed, the  people  at  the  top  of the  hierarchy  live  longer  than  the  people  at  the  bottom. But  that  there  was  a  strong  correlation between  the  social  hierarchy  and  death  rates. The  administers,  the  administrators had  the  lowest  death  rates, followed  by  the  professional  and  executive  staff, followed  by  the  clerical  staff, followed  by  the  ancillary  staff. In  other  words,  the  health  of these  men  followed  a  social  gradient. The  researchers  looked  at  one  disease  in more  detail,  coronary  heart  disease. They  wanted  to  see  if  they  could explain  why  this  was  happening. They  found  that  one-third  of  the  social  gradient  in coronary  heart  disease  mortality was  attributable  to  smoking. Hi,  plasma,  cholesterol,  high  blood  pressure, being  overweight,  and  having low  levels  of  physical  activity. So  some  of  the  difference  was  accounted for  by  differences  in  lifestyle. People  further  down  the  social  hierarchy were  more  likely  to  smoke, less  likely  to  take  physical  activity, and  more  likely  to  be  overweight than  people  higher  up  the  social  hierarchy. However,  that  left  two-thirds  of the  social  gradient  in coronary  heart  disease  mortality,  unexplained. Even  after  taking  lifestyle  factors  into  account. Death  rates  from  coronary  heart  disease  were inversely  related  to  social  position. We  still  don't  fully  understand  why  this  is  the  case. One  possible  explanation  is  that  chronic  stress, which  impacts  disproportionately  on  poorer  people, might  be  to  blame. However,  more  research  is  needed  in  this  important  area. I  want  now  to  look  briefly  at  how cohort  studies  are  analysed. You  can  see  that  at  the  end  of  a  cohort  study, there  are  four  potential  groups. Groupie  is  the  exposed  group who  developed  the  outcome  of  interest. E.g.  smokers  who  have  developed  lung  cancer. Group  B.  Group  B  is the  exposed  group  who  did not  develop  the  outcome  of  interest. So  smokers  who  don't  have  lung  cancer. Group  C  is  the  unexposed  group that  developed  the  outcome  of  interest. That  would  be  non-smokers  who  developed  lung  cancer. And  finally,  Group  D  is the  unexposed  group  that did  not  develop  the  outcome  of  interest. So  non-smokers  who  don't  have  lung  cancer. This  information  can  be transferred  onto  a  two-by-two  table, like  the  one  on  the  slide. The  first  thing  you  might  do  in a  cohort  study  is  to  examine the  incidence  of  the  outcome in  the  exposed  population,  e.g. the  incidence  of  lung  cancer  among  smokers. This  is  calculated  by  dividing  the  number  of  smokers with  lung  cancer  by  the  total  number  of  smokers. So  a  divided  by  a  plus  b. In  other  words,  this  is  the  proportion of  the  exposed  participants. He  went  on  to  develop the  outcome  during  the  period  of  the  study. Next,  you  can  work  out  the  incidence  of the  outcome  in  the  unexposed  population. In  other  words,  the  incidence  of lung  cancer  in  non-smokers. This  is  calculated  by  the  number  of lung  cancer  cases  who  are  non-smokers, divided  by  the  total  number  of  non-smokers. So  c  divided  by  c  plus  d. Or  in  other  words,  the  proportion  of unexposed  participants  who  went  on  to develop  the  outcome  during  the  period  of  the  study. At  this  point,  I  want  to  introduce  a  new  metric, relative  risk,  also  referred  to  as  the  risk  ratio. Relative  risk  is  the  main  outcome  indicator produced  by  cohort  studies. Relative  risk  is  defined  as the  incidence  of  the  outcome  in the  exposed  population  divided  by the  incidence  of  the  outcome  in  the  unexposed  population. So  in  our  example, the  incidence  of  lung  cancer  in  smokers divided  by  the  incidence  of  lung  cancer  in  non-smokers. It  is  calculated  as  shown  on  the  slide. It's  a  measure  of  the  association between  the  exposure  and  the  outcome. So  how  do  you  interpret  relative  risk? You  can  see  from  the  equation  that  if the  incidence  in  the  exposed  population  is  exactly  the same  as  the  incidence  in the  unexposed  population  than the  relative  risk  will  be  equal  to  one. So  if  you  get  a  relative  risk  of  exactly  one, it  indicates  that  there  is no  association  between  the  exposure  and  the  outcome. If  the  incidence  in  the  exposed  population, population  is  greater  than the  incidence  in  the  unexposed  population, the  relative  risk  will  be  greater  than  one. A  relative  risk  group  greater  than  one  indicates that  the  exposure  is  a  risk  factor  for  the  outcome. In  other  words,  that  there's a  positive  association  between the  exposure  and  the  outcome. Conversely,  if  the  incidence  in the  exposed  population  is  less than  the  incidence  in  the  unexposed  population, the  relative  risk  will  be  less  than  one. A  relative  risk  less  than  one indicates  that  the  exposure  is protective  against  the  outcome. Or  in  other  words,  that  there's a  negative  association  between the  exposure  and  the  outcome. This  latter  situation  might  occur,  e.g. if  the  exposure  is  a  vaccine  where  you  might  expect it  to  protect  against  the  development  of  disease. Like  any  study  design. Cohort  studies  have  their  advantages  and  disadvantages. Let's  start  with  the  advantages. Cohort  studies  are  particularly  good  for  rare  exposures. This  is  because  you  select  your  study  sample  on the  basis  of  the  exposure.  If  e.g. you  wanted  to  investigate  the  impact  of  exposure  to a  particular  pesticide  and the  risk  of  Parkinson's  disease. You  would  set  out  to  recruit  people  known to  have  been  exposed  to  that  pesticide. Another  advantage  is  that cohort  studies  enable  you  to  study  multiple  outcomes. So  in  the  example  I've  just  given. You  could  study  the  association  of pesticide  exposure  to  a  whole  range of  different  disease  outcomes. Another  advantage  of  cohort  studies, as  I  discussed  in  the  introduction  to  this  presentation, is  that  they  can  provide  data  on the  temporal  relationship  between the  exposure  and  the  outcome. In  contrast  to  cross-sectional  studies, you  can  be  sure  that  the  exposure  was present  before  the  outcome  occurred. Compared  to  case  control  studies, which  we  will  look  at  next. There's  a  lower  risk  of  selection  bias  in  cohort  studies. This  is  particularly  true  for  prospective  cohort  studies. The  final  advantage  of  cohort  studies  is  that  you  can estimate  the  incidence  of  disease  as  we  have  seen. However,  cohort  studies  have  disadvantages  too. They  are  inefficient  for  rare  outcomes. If  it  is  a  rare  disease, you  have  to  wait  a  long  time  or  have a  very  large  study  to  develop  enough  outcomes. Cohort  studies  are  relatively expensive  and  time-consuming, particularly  prospective  cohort  studies. Finally,  cohort  studies  have a  high  risk  of  loss  to  follow  up. In  other  words,  people  being  recruited  into the  study  and  then  later  dropping  out. This  is  a  really  serious  problem  because  it can  distort  and  bias  the  results  of  the  study. If  the  dropouts  differ  in important  respects  to  those  who  remained  in  the  study. This  is  the  final  slide  of  the  presentation. What  should  you  look  out  for  when critically  appraising  a  cohort  study? The  points  on  this  slide  are summarised  from  the  Joanna  Briggs  Institute and  critical  appraisal  skills programme,  critical  appraisal  tools. Firstly,  pay  attention  to  how  participants  were  selected. The  two  groups,  the  exposed  and  the  unexposed, should  be,  should  be  recruited  from  the  same  population. They  should  be  as  alike  as  possible, except  for  the  exposure  of  interest. Think  about  the  generalizability of  the  results  of  the  cohort  studies. Cohort  studies  often  use occupational  groups  as  their  study  population. But  this  limits  the  generalizability of  the  results  to  the  study, to  the  results  of  the  study  to  the  general  population. Secondly,  pay  attention  to  how the  exposure  and  the  outcome  were  measured. This  should  be  done  objectively,  not  subjectively. Any  tools  should  ideally  be  validated. E.g.  if  the  outcome  is  depression, a  validated  tool  to  measure  depression  is  much  more reliable  than  relying  on self-reported  depressive  symptoms. It's,  Is  it  clear  that participants  were  free  of the  outcome  at  the  start  of  the  study, because  this  is  a  really  important  feature of  a  cohort  study. So  how  do  you  know  that? What  is  there  in  the  study that  demonstrates  that  participants were  indeed  free  at the  outcome  at  the  start  of  the  study. Another  important  point  about  the  measurement of  exposures  and  outcomes  is  that the  same  approach  should  be  used  in both  groups  to  avoid  measurement  bias. So  the  exposed  and  the unexposed  should  be  treated  exactly  the same  in  how  they  are measured  and  how  they  are followed  up  throughout  the  study. Next  half, potential  confounding  factors  been identified  and  measured  in  this  study, participants  are  really  important  confounding  factors missing  have  confirmed  the  factors being  taken  into  account  in  the  analysis  of  the  results. So  in  the  analysis  section, look  out  for  where  the  results  have  been adjusted  for  confounding  factors, whether  regression  methods  have been  used  in  the  analysis, whether  sensitivity  analysis  have  been  done  and  so  on. Pay  attention  to  how  participants  were  followed  up. What's  the  follow-up  time  reported? And  was  it  long  enough  for  the  outcomes  to occur  where  all  participants followed  up  or  were  there  dropouts  from  the  study, which  is  called  loss  to  follow  up. If  there  was  lost  to  follow  up, are  the  reasons  for  the  loss  to follow-up  described  and  explored. Was  there  anything  special  or  different  about  the  people leaving  the  study  compared  to  the  people  who  stayed, and  what  strategies  were  used  to  maximise  follow-up. Those  are  the  key  elements  that  you  need  to  take into  account  when  you're critically  appraising  a  cohort  study. But  you  should  always  use a  defined  tool  to  do  your,  your  critical  appraisal. And  the  Joanna  Briggs  Institute  and  the  CASP  skills programme  tool  are  two  suggestions  that  you  can  use.