Public

 Health  

 
PUBLIC  HEALTH  RESEARCH  
Observa5onal  Research  Design  in  
Epidemiology  
 
Dr  Rui  Bap*sta-­‐Gonçalves  
 Learning  Outcomes  
•  To  recognise  the  different  types  of  
epidemiological  research  design    
•  To  be  aware  of  the  applicability  of  
the  different  designs  and  its  
strengths  and  weaknesses  
 
Röhrig,  B;  Prel,  J  d;  Wachtlin,  D;  BleJner,  M  
Types  of  Study  in  Medical  Research—Part  3  of  a  Series  on  Evalua5on  of  Scien5fic  Publica5ons  
Dtsch  Arztebl  Int  2009;  106(15):  262-­‐8;  DOI:  10.3238/arztebl.2009.0262  
Röhrig,  B;  Prel,  J  d;  Wachtlin,  D;  BleJner,  M  
Types  of  Study  in  Medical  Research—Part  3  of  a  Series  on  Evalua5on  of  Scien5fic  Publica5ons  
Dtsch  Arztebl  Int  2009;  106(15):  262-­‐8;  DOI:  10.3238/arztebl.2009.0262  
 Research  Design  
OBSERVATIONAL   INTERVENTIONAL  
Examines  associa*on   Explores  the  associa*on  
between  risk  factors   between  interven*ons  
and  outcomes,  trying  to  
prove  that  A  causes  B  

ANALYTICAL  
Provides  informa*on  on  
associa*on  between  
exposure  and  risk  factors  

DESCRIPTIVE  
Provides  informa*on  on  
frequency  and  distribu*on  
 Common  Research  Design  
 
•  Cross-­‐Sec5onal    
•  Cohort  
•  Case-­‐control  

•  (Experimental:  RCTs,  clinical  trials)  

 
 Cross  Sec*onal  Studies  
•  Snapshots  à  big  picture  
•  Captures  informa*on  pertaining  to  
disease  or  exposure  variables  (or  both)  at  
one  point  in  *me  
•  Descrip*ve  
•  Role  in  PH  :  Preven*on  

 
 Sample  

•  Representa*ve  of  the  popula*on  to  

which  you  wish  to  apply  the  findings  
from  the  research  
•  Of  sufficient  sample  size  to  extrapolate  
your  findings  to  the  popula*on  
•  Must  consider  *me,  money  and  
resources  
Popula5on   Sample  

All  physiotherapists  working   50  physios  

in  the  NHS  
VARIABLES:  
 
Diabe*c  pa*ents  in  London   40  diabe*c  pa*ents  with   Dura*on,  Type,  
with  peripheral  neuropathy   peripheral  neuropathy  from   socioeconomic,  age,  
Wandsworth  PCT   weight,  gender….  

UK  Teenage  mothers   30  young  mothers  aJending    

child  clinic  
 Data  collec*on  &  Analysis  

•  Exposure  to  risk  factors  

•  Outcomes  
•  Associa*on  between  risk  factors  and  
outcomes  
•  Es*ma*on  of  prevalence  
•  (some*mes  called  prevalence  studies)  

 
 Strengths    
•  Inexpensive  and  *me-­‐friendly  

•  Numerous  factors  of  assessment  

•  Public  health  planning:  disease  ae*ology  and  

genera*on  of  hypothesis  

 
 Weaknesses  
•  NO  causality:  cannot  link  the  exposure  and  the  
disease  
•  No  discrimina*on  between  dura*on  
•  Cannot  predict  future  health  events  
•  Bias  

 
 [Bias]  
•  “the  introduc*on  of  error  that  produces  devia*ons  
or  distor*ons  in  the  data  that  are  predominantly  one  
direc*on,  as  opposed  to  random  error.  “  Jekel,  Katz  
&  Elmore  (2001)  
•  Selec5on  bias:  introduc5on  of  bias  by  ini5al  
differences  between  groups  (threatens  int  validity)  
•  Length  bias:  milder  cases  of  disease  are  detected  
dispropor5onally  in  popula5on  screening  
programmes  

 
 Cross  Sec*onal  
•  A  cross  sec*onal  survey  was  carried  out  among  a  mul*racial  
workforce  at  worksites  in  New  Zealand  by  Scragg  and  colleagues  
between  1988  and  1990.  The  survey  studied  5677  staff  aged  
40-­‐64  years.  The  subjects  were  asked  about  their  age,  ethnicity,  
past  medical  history,  occupa*on  and  income.  Their  height,  
weight  and  blood  pressure  were  recorded  and  an  oral  glucose  
tolerance  test  to  detect  diabetes  was  performed.  The  study  
showed  that  the  prevalence  of  diabetes  increased  with  age,  was  
more  common  in  Maoris  and  that  approx  50%  of  workers  with  
diabetes  were  previously  undiagnosed.  The  prevalence  of  
diabetes  was  also  significantly  correlated  with  weight  and  low  
income.  
Cross  Sec*onal  
 Ac*vity  
•  Is  this  a  descrip*ve  or  an  analy*cal  study  
•  Does  it  maJer  that  data  from  some  of  the  
eligible  children  were  not  included  in  the  
analysis?  
•  Describe  the  results  show  in  table  1.1  
 Cohort  studies  
•  A  study  following  2  or  more  groups  from  exposure  to  
outcome  
•  Concurrent  cohort  study  

•  Retrospec5ve  cohort  study  

•  Ambidirec5onal  cohort  study  

 
“the  inves*gator  selects  a  group  of  
exposed  individuals  and  a  group  of  non-­‐
exposed  individuals  and  follows  up  both  
groups  to  compare  the  incidence  of  
disease  (or  rate  of  death  from  disease)  in  
the  two  groups.”  
 
Gordis  (2004)  
 Defined  popula*on  

Non-­‐randomisa*on  

Exposed   Non-­‐exposed  

disease   No  disease  
disease   No  disease  
 DISEASE   NO  DISEASE   INCIDENCE  RATE  

EXPOSED   a   b   a/a+b  

NON  EXPOSED   c   d   c/c+d  

 Strengths  
•  Treatment  is  not  withheld  from  subjects,  and  they  
are  not  ar*ficially  subjected  to  poten*al  hazards  
•  Subjects  matched  for  possible  confounders  
•  Can  obtain  informa*on  on  one  exposure  and  
mul*ple  diseases  
•  Can  obtain  informa*on  on  persons  who  change  
exposure  status  
•  Best  design  to  establish  incidence  and  natural  
disease  history  
•  Helps  calculate  incidence  rates  and  rela*ve  risk    

 
 Weaknesses  
•  Difficult  and  expensive  
•  Confounding  –  exposure  may  be  related  to  
something  unknown  
•  Difficult  to  obtain  control  group  is  therapy  is  popular  
or  most  people  have  been  exposed  
•  Altered  behaviours  may  affect  disease  
•  Rare  diseases  are  difficult  to  study  
•  Loss  to  follow  up  

 
 Cohort  
•  Mortality  in  rela*on  to  smoking:  40  years’  
observa*ons  on  male  Bri*sh  Doctors  (Doll  et  al,  
1994  BMJ)  
•  Yielded  two  observa*ons  that  could  not  have  
been  made  from  descrip*ve  studies  alone:  
Ø Sequence  of  events  
Ø Dose-­‐response  effect   Causal  hypothesis    
 Case-­‐control  Studies  
•  “A  case-­‐control  study  is  an  inquiry  in  which  
groups  of  individuals  are  selected  based  on  
whether  they  do  (cases)  or  they  do  not  (the  
controls)  have  the  disease  of  which  the  
ae<ology  is  to  be  studied.  The  two  groups  are  
then  used  to  evaluate  the  rela<on  to  the  study  
disease  of  exis<ng  of  past  characteris<cs  
among  the  individuals.”  

     MacMahon  &  Trichopoulous  (1996)  

 
 Case-­‐control  

•  A  study  design  to  help  establish  if  an  exposure  

is  associated  with  an  outcome  
•  Always  retrospec*ve  –  looking  to  see  who  had  
exposures,  compared  between  case  and  
control  groups  
•  Cases:  a  group  know  to  have  the  outcome  
•  Controls:  a  group  known  to  be  without  the  
outcome  (representa*ve!)  

 
In  the  1940s,  Sir  Norman  Gregg,  an  Australian  
ophthalmologist,  observed  a  number  of  infants  and  
young  children  in  his  prac*ce  who  presented  with  an  
unusual  form  of  cataract.  
He  noted  that  these  children  had  been  in  utero  during  
the  *me  of  a  rubella  (German  measles)  outbreak.  He  
suggested  that  there  was  an  associated  between  
prenatal  rubella  exposure  and  the  development  of  
unusual  cataracts.”  
         Gordis  (2004,  p.159)  
 Steps    
•  Define  cases  and  controls  (be  very  clear!!!)  
•  Sampling  
•  Data  collec5on  
•  Recruit  two  controls  for  every  case  
•  Controls  must  not  have  the  disease  being  
inves5gated  at  the  5me  the  case  arose  (index  day).  

•  Odds  ra5ons  and  confidence  intervals  

 
 Procedure  

Data  can  be  obtained  by  interview,  ques*onnaire  or  

reference  to  pre-­‐exis*ng  records  or  a  combina*on  of  
one  or  more  of  these  sources.    
 
Same  method  used  for  both  cases  and  controls  
 
 Strengths  

•  Ethical  way  of  inves*ga*ng  associated  (i.e.  risk  

factors  for  rare  disease)  
•  Quick  and  inexpensive  with  evident  clinical  
applica*on  (does  not  have  to  wait  for  disease  
to  develop,  like  what  happens  with  cohorts)  
•  Data  collec*on  rela*vely  easy  and  available  
•  Provides  support  for  the  analysis  of  
hypothesis  

 
 Weaknesses  

•  Weak  data  
•  Non-­‐homogeneous  cases  can  be  hard  to  
analyse  
•  Does  not  provide  incidence  rates  as  it  is  not  
based  on  defined  popula*ons  
•  Recall  bias  (retrospec*ve  studies…)  
•  Difficul*es  in  recrui*ng  controls  

 
 Case  control  
SlaJery,  et  al  (1989)  Am  J  Epidem  130  
 
•  Sexual  ac*vity,  contracep*ve  method,  genital  infec*ons  and  cervical  cancer  
•  Between  1984  and  1987  in  Utah  (Mormons)  
•  Explore  the  rela*onship  between  cervical  cancer  and  sexual  ac*vity,  the  use  of  
barrier  methods  of  contracep*on  and  certain  types  of  genital  infec*on.  
•  Subjects:  20-­‐59y  women,  newly  diagnosed  with  cervical  cancer  
•  Controls  iden*fied  by  use  of  a  random  digit  dialling    technique  and  matched  to  
cases  by  5-­‐year  age  intervals  
•  Several  risk  factors  iden*fied:  mul*ple  sexual  partners,  current  mate  having  
mul*ple  sexual  partners,  reported  Trichomonas  infec*on  and  serological  
evidence  of  herpes  virus  type  2  infec*on  
•  Protec*ve  effect  was  noted  from  use  of  diaphragm  or  condoms  in  women  who  
reported  more  than  one  sexual  partner  
•  Supports  the  hypothesis  that  cervical  cancer  is  due  to  a  sexually  transmiJed  
agent