How to cra the "Approach" sec1on of an R grant applica1on

Page created by Emma Mccormick

Hobbies & Interests

English

Like
Share
Embed
Fullscreen
Slides
Download HTML
Download PDF
Abuse

←

→

Page content transcription

If your browser does not render page correctly, please read the page content below

How to cra the "Approach" sec1on of an R grant applica1on

How	
  to	
  cra)	
  the	
  “Approach”	
  
sec1on	
  of	
  an	
  R	
  grant	
  applica1on	
  
                    David	
  Elashoﬀ,	
  PhD	
  
           Professor	
  of	
  Medicine	
  and	
  Biosta7s7cs	
  
      Director,	
  Department	
  of	
  Medicine	
  Sta7s7cs	
  Core	
  
             Leader,	
  CTSI	
  Biosta7s7cs	
  Program

Overview	
  
•   Preliminary	
  Data	
  
•   Study	
  Design	
  
•   Sample	
  Size	
  and	
  Power	
  Analysis	
  
•   Sta7s7cal	
  Methods	
  
•   Collaborators	
  
•   Wri7ng	
  Strategies

Preliminary	
  Data	
  
• Primary	
  Ques7on:	
  “Is	
  there	
  reason	
  to	
  believe	
  
     that	
  the	
  study	
  hypotheses	
  could	
  be	
  true	
  and	
  is	
  
     this	
  research	
  team	
  capable	
  of	
  carrying	
  out	
  the	
  
     study?”

Necessary	
  Elements:	
  Preliminary	
  
                  Data	
  
• Strong	
  and	
  relevant	
  preliminary	
  data	
  key	
  for	
  
  R01	
  grants	
  
• Demonstrate:	
  
    – Exper7se	
  with	
  assays	
  
    – Novel	
  assays	
  work	
  in	
  pa7ents/samples	
  to	
  be	
  
      collected	
  
    – Support	
  for	
  hypotheses	
  
• Use	
  ﬁgures	
  and	
  tables	
  where	
  possible

Ways	
  to	
  Fail:	
  Preliminary	
  Data	
  
•   Insuﬃcient	
  annota7on	
  for	
  ﬁgures/tables	
  
•   Poor	
  data	
  analy7c	
  techniques	
  
•   Weak	
  support	
  for	
  hypotheses	
  
•   Unrealis7cally	
  strong/naïve	
  preliminary	
  results	
  
•   Presen7ng	
  needle	
  in	
  a	
  haystack	
  results	
  
•   Presen7ng	
  too	
  much	
  preliminary	
  data	
  at	
  
    expense	
  of	
  rest	
  of	
  the	
  approach

Study	
  Design	
  
• Primary	
  Ques7on:	
  “Is	
  the	
  design	
  of	
  the	
  study	
  
  appropriate	
  to	
  address	
  the	
  study	
  aims?”

Necessary	
  Elements:	
  Study	
  Design	
  
• What	
  is	
  overall	
  study	
  design	
  (RCT,	
  Cohort	
  
  study,	
  Case-‐Control,	
  Cross-‐sec7onal,	
  
  Biomarkers)	
  	
  
• Describe	
  endpoints	
  and	
  clarify,	
  if	
  necessary,	
  
  how	
  they	
  will	
  be	
  quan7ﬁed	
  and	
  their	
  
  measurement	
  scale.	
  
• Describe	
  study	
  popula7on	
  and	
  control	
  groups	
  
• Inclusion/Exclusion	
  Criteria	
  
• Describe	
  all	
  study	
  measures	
  with	
  appropriate	
  
  measurement	
  process	
  details

Addi7onal	
  Considera7ons:	
  Study	
  
                     Design	
  
• Describe	
  exis7ng	
  popula7on	
  clearly.	
  
  	
  -‐	
  Include	
  relevant	
  demographics	
  
  	
  -‐	
  Include	
  informa7on	
  on	
  prognos7c	
  or	
  
               	
  confounding	
  measures.	
  

• Nothing	
  says	
  that	
  this	
  is	
  a	
  ready	
  to	
  go	
  study	
  
  be^er	
  than	
  a	
  clearly	
  deﬁned	
  popula7on	
  that	
  is	
  
  relevant	
  to	
  the	
  study	
  aims.

Addi7onal	
  Considera7ons	
  
•   Randomiza7on	
  methods	
  for	
  clinical	
  trials	
  
•   Collect	
  confounding	
  factors	
  
•   How	
  long	
  will	
  follow-‐up	
  period	
  be?	
  
•   Validity	
  and	
  reliability	
  of	
  study	
  measures	
  
•   Subject	
  matching?	
  
•   Valida7on	
  of	
  model	
  building	
  either	
  with	
  cross-‐
    valida7on	
  or	
  training-‐test	
  designs

Ways	
  to	
  Fail:	
  Study	
  Design	
  
• Study	
  popula7on	
  or	
  design	
  doesn’t	
  match	
  
  objec7ves	
  
• Insuﬃcient	
  7me	
  for	
  recruitment	
  and	
  follow-‐
  up.	
  
• Lack	
  of	
  clarity	
  with	
  respect	
  to	
  availability	
  of	
  
  subjects	
  
• Very	
  uninteres7ng	
  to	
  read	
  technical	
  details	
  of	
  
  assays	
  that	
  are	
  standard

Sample	
  Size	
  
• Primary	
  Ques7on:	
  “Is	
  the	
  sample	
  size	
  
  suﬃcient	
  to	
  give	
  the	
  study	
  the	
  ability	
  to	
  
  answer	
  the	
  primary	
  study	
  ques7ons?”

Necessary	
  Elements:	
  Sample	
  Size	
  
• Iden7fy	
  study	
  endpoint(s)	
  for	
  all	
  aims.	
  
• Clearly	
  describe	
  sample	
  size	
  for	
  each	
  aim	
  
• For	
  each	
  endpoint:	
  
    – What	
  is	
  the	
  eﬀect	
  of	
  interven7on	
  or	
  magnitude	
  of	
  the	
  
      rela7onship?	
  
    – How	
  much	
  variability?	
  
    – Level	
  of	
  power?	
  
    – One	
  or	
  two	
  sided	
  test?	
  
    – What	
  is	
  the	
  sta7s7cal	
  test	
  used	
  to	
  compute	
  power?

Addi7onal	
  Considera7ons:	
  Sample	
  Size	
  
• Account	
  for	
  study	
  dropouts	
  
• Account	
  for	
  mul7ple	
  comparisons	
  (either	
  
  Bonferroni	
  or	
  False	
  Discovery	
  Rate)	
  
• Ocen	
  useful	
  to	
  examine	
  sample	
  sizes	
  for	
  a	
  
  variety	
  of	
  scenarios	
  when	
  uncertainty	
  exists	
  
  concerning	
  what	
  is	
  to	
  be	
  expected	
  for	
  an	
  
  endpoint

Ways	
  to	
  Fail:	
  Sample	
  Size	
  
• No	
  power	
  analysis	
  
• Sample	
  size	
  calcula7on	
  does	
  not	
  have	
  suﬃcient	
  
  informa7on	
  for	
  a	
  reviewer	
  to	
  replicate	
  
• Sample	
  size	
  calcula7on	
  does	
  not	
  use	
  relevant	
  
  preliminary	
  data	
  or	
  methods	
  described	
  in	
  the	
  
  sta7s7cal	
  analysis	
  sec7on.	
  
• Predic7on	
  modeling	
  with	
  large	
  number	
  of	
  
  predictors	
  rela7ve	
  to	
  sample	
  size	
  
• Unrealis7c	
  assump7ons	
  about	
  magnitude	
  of	
  
  eﬀect

Bad	
  Examples	
  
“A	
  previous	
  study	
  in	
  this	
  area	
  recruited	
  150	
  subjects	
  and	
  found	
  highly	
  signiﬁcant	
  
results	
  (p=0.014),	
  and	
  therefore	
  a	
  similar	
  sample	
  size	
  should	
  be	
  suﬃcient	
  here.”	
  
	
  
“Our	
  lab	
  usually	
  uses	
  10	
  mice	
  per	
  group.”	
  
	
  
“Sample	
  sizes	
  are	
  not	
  provided	
  because	
  there	
  is	
  no	
  prior	
  informa7on	
  on	
  which	
  to	
  
base	
  them.”	
  
	
  
"The	
  throughput	
  of	
  the	
  clinic	
  is	
  around	
  50	
  pa7ents	
  a	
  year,	
  of	
  whom	
  10%	
  may	
  refuse	
  
to	
  take	
  part	
  in	
  the	
  study.	
  Therefore	
  over	
  the	
  2	
  years	
  of	
  the	
  study,	
  the	
  sample	
  size	
  
will	
  be	
  90	
  pa7ents.	
  “	
  
	
  
“It	
  is	
  es7mated	
  that	
  for	
  a	
  sample	
  size	
  consis7ng	
  of	
  6	
  animals	
  in	
  each	
  trial	
  and	
  with	
  a	
  
tumor	
  volume	
  variance	
  from	
  0.1	
  to	
  1.0	
  cm3	
  –	
  that	
  when	
  the	
  diﬀerence	
  in	
  the	
  
popula7on	
  reaches	
  0.25,	
  the	
  power	
  will	
  reach	
  100%.”

Good	
  Examples	
  
“A	
  sample	
  size	
  of	
  38	
  in	
  each	
  group	
  will	
  be	
  suﬃcient	
  to	
  detect	
  a	
  diﬀerence	
  of	
  5	
  points	
  
on	
  the	
  Beck	
  scale	
  of	
  suicidal	
  idea7on,	
  assuming	
  a	
  standard	
  devia7on	
  of	
  7.7	
  points,	
  a	
  
power	
  of	
  80%,	
  assuming	
  	
  a	
  two	
  sided	
  signiﬁcance	
  level	
  of	
  5%	
  and	
  a	
  two	
  sample	
  t-‐test.	
  
This	
  number	
  has	
  been	
  increased	
  to	
  60	
  per	
  group	
  (total	
  of	
  120),	
  to	
  allow	
  for	
  a	
  predicted	
  
drop-‐out	
  from	
  treatment	
  of	
  around	
  one	
  third.	
  This	
  diﬀerence	
  of	
  5	
  points	
  is	
  based	
  on	
  
our	
  prior	
  study	
  in	
  which…..	
  ”	
  
	
  
“A	
  sample	
  size	
  of	
  292	
  babies	
  (146	
  in	
  each	
  of	
  the	
  treatment	
  and	
  placebo	
  groups)	
  will	
  be	
  
suﬃcient	
  to	
  detect	
  a	
  diﬀerence	
  of	
  16%	
  between	
  groups	
  in	
  the	
  sepsis	
  rate	
  at	
  14	
  days,	
  
with	
  80%	
  power.	
  This	
  16%	
  diﬀerence	
  represents	
  the	
  diﬀerence	
  between	
  a	
  50%	
  sepsis	
  
rate	
  in	
  the	
  placebo	
  group	
  and	
  a	
  34%	
  rate	
  in	
  the	
  treatment	
  group.	
  This	
  assumes	
  a	
  Chi-‐
square	
  test	
  with	
  a	
  two	
  sided	
  0.05	
  signiﬁcance	
  level.	
  This	
  es7mated	
  diﬀerence	
  in	
  sepsis	
  
rate	
  is	
  based	
  on	
  the	
  study	
  of	
  Bob	
  et	
  al	
  [ref]	
  in	
  which	
  they	
  observed….”

Sta7s7cal	
  Methods	
  
• Primary	
  Ques7on:	
  “Are	
  the	
  sta7s7cal	
  methods	
  
  appropriate	
  for	
  the	
  analysis	
  of	
  the	
  data	
  that	
  
  will	
  be	
  collected?”

Necessary	
  Elements:	
  Sta7s7cal	
  
                    Methods	
  
• Need	
  methods	
  sec7on	
  for	
  each	
  aim.	
  
• Clearly	
  describe	
  analy7c	
  strategies	
  for	
  each	
  
  endpoint.	
  
• Methods	
  should	
  be	
  appropriate	
  for	
  type	
  of	
  
  variable	
  (ex.	
  categorical,	
  ordinal,	
  count)	
  and	
  
  study	
  design	
  
• Typically	
  includes	
  inferen7al	
  tes7ng	
  of	
  
  endpoints	
  and	
  model	
  building

Addi7onal	
  Considera7ons:	
  Sta7s7cal	
  
                 Methods	
  
• Sta7s7cal	
  methods	
  appropriate	
  for	
  sample	
  
  size	
  (ex.	
  Fisher	
  test	
  vs	
  Chi-‐square	
  test)	
  
• Include	
  evalua7on	
  and	
  valida7on	
  strategies	
  for	
  
  regression/predic7on	
  models	
  
• Can	
  include	
  model	
  assump7on	
  checking	
  
  methods	
  
• Accoun7ng	
  for	
  missing	
  data

Ways	
  to	
  Fail:	
  Sta7s7cal	
  Methods	
  
• Ignoring	
  key	
  confounders	
  or	
  demographic	
  
  variables.	
  
• Ignoring	
  standard	
  prognos7c	
  or	
  predic7ve	
  
  measures	
  in	
  models	
  
• Describing	
  socware	
  but	
  not	
  ideas/methods	
  
• Analy7cal	
  approach	
  not	
  appropriate	
  for	
  design	
  
  and	
  research	
  ques7on

Ways	
  to	
  Fail:	
  Sta7s7cal	
  Methods	
  
• Ignoring	
  key	
  confounders	
  or	
  demographic	
  
  variables.	
  
• Ignoring	
  standard	
  prognos7c	
  or	
  predic7ve	
  
  measures	
  in	
  models	
  
• Describing	
  socware	
  but	
  not	
  ideas/methods	
  
• Analy7cal	
  approach	
  not	
  appropriate	
  for	
  design	
  
  and	
  research	
  ques7on

Collaborators	
  
• Primary	
  Ques7on:	
  “Does	
  the	
  study	
  have	
  
  appropriate	
  collaborators	
  with	
  suﬃcient	
  eﬀort	
  
  to	
  perform	
  the	
  research	
  described?”

Necessary	
  Elements:	
  Collaborators	
  
• Need	
  an	
  iden7ﬁed	
  sta7s7cal	
  collaborator	
  with	
  
  appropriate	
  experience	
  
• Biosketchs	
  for	
  faculty	
  collaborator	
  
• Budget	
  jus7ﬁca7on	
  for	
  collaborator	
  
• Le^er	
  of	
  Support	
  if	
  no	
  funding	
  is	
  in	
  
  applica7on.	
  
   – Make	
  use	
  of	
  collaborators	
  from	
  on-‐campus	
  service	
  
     groups.	
  (Ex:	
  CTSI,	
  Cancer	
  Center)

Addi7onal	
  Considera7ons:	
  
                Collaborators	
  	
  
• Staﬀ	
  collaborator	
  only	
  need	
  biosketch	
  if	
  no	
  
  faculty	
  on	
  applica7on.	
  
• Can	
  include	
  small	
  %	
  eﬀort	
  for	
  expensive	
  
  faculty	
  and	
  larger	
  %	
  for	
  staﬀ	
  support.	
  
• Make	
  sure	
  areas	
  of	
  weakness	
  are	
  covered	
  with	
  
  experienced	
  collaborator	
  
• Don’t	
  include	
  many	
  collaborators	
  with	
  
  minimal	
  eﬀort	
  
• Not	
  enough	
  to	
  men7on	
  collaborators	
  and	
  
  write	
  that	
  they	
  will	
  take	
  care	
  of	
  details

Wri7ng	
  Strategies	
  
• Use	
  the	
  resources	
  and	
  human	
  subjects	
  
  sec7ons	
  to	
  full	
  eﬀect	
  
    – Can	
  give	
  details	
  of	
  available	
  study	
  popula7on	
  and	
  
      subject	
  demographics	
  
• Standard	
  experimental	
  methods	
  can	
  be	
  
  referenced	
  
• Long	
  blocks	
  of	
  text	
  are	
  boring	
  a	
  can	
  ocen	
  get	
  
  skimmed.	
  
• Emphasize	
  key	
  points:	
  bold,	
  underline

Wri7ng	
  Strategies	
  
• Graphical	
  displays:	
  
   – Theore7cal	
  Framework	
  
   – Experimental	
  Design	
  
   – Aims	
  ﬂowchart	
  
   – Pa7ent	
  characteris7cs	
  
   – Study	
  measures

Don’t	
  Waste	
  Space

Grant	
  Applica7ons	
  Assistance	
  
• Assistance	
  with	
  preparing	
  grant	
  applica7ons	
  
  (CTSI)	
  
    –   Study	
  Design	
  
    –   Data	
  Analysis	
  Protocols	
  
    –   Sample	
  Size	
  and	
  Power	
  Analysis	
  
    –   Budge7ng	
  and	
  Iden7fying	
  Appropriate	
  Collaborators	
  
    –   Core	
  facili7es	
  	
  
• Substan7al	
  lead	
  7me	
  with	
  opportunity	
  for	
  
  mul7ple	
  itera7ons	
  is	
  necessary	
  for	
  high	
  quality	
  
  grant	
  applica7on	
  assistance:	
  Study	
  Design	
  vs	
  
  Analysis	
  sec7ons

Final	
  Thoughts	
  
• Consult	
  sta7s7cal	
  collaborator	
  for	
  study	
  design	
  
  and	
  approximate	
  sample	
  size	
  some	
  weeks	
  in	
  
  advance	
  
• Most	
  successful	
  proposals	
  require	
  mul7ple	
  
  itera7ons	
  of	
  research	
  design	
  sec7ons

You can also read