• No results found

The Development of a Clinical Prediction Tool to Support Clinicians in the Assessment of the Risks of Fetal Asphyxia and Failure to Progress in Term Pregnancies

N/A
N/A
Protected

Academic year: 2021

Share "The Development of a Clinical Prediction Tool to Support Clinicians in the Assessment of the Risks of Fetal Asphyxia and Failure to Progress in Term Pregnancies"

Copied!
50
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The  Development  of  a  Clinical  Prediction  

Tool  to  Support  Clinicians  in  the  

Assessment  of  the  Risks  of  Fetal  Asphyxia  

and  Failure  to  Progress  in  Term  

Pregnancies  

 

 

 

 

 

 

 

Uloma  C  Ogba  

April  2015  

(2)

The  Development  of  a  Clinical  Prediction  

Tool  to  Support  Clinicians  in  the  

Assessment  of  the  Risks  of  Fetal  Asphyxia  

and  Failure  to  Progress  in  Term  

Pregnancies  

 

Student   Uloma  C  Ogba   Collegekaart  nummer:  10021671   E-­‐mail:  u.ogba@amc.uva.nl     Mentor   Sabine  Ensing,  MD   AMC  

Obstetrics  and  Gynecology/Department  of  Medical  Informatics   s.ensing@amc.uva.nl  

  Tutor  

Anita  Ravelli,  PhD  

Department  of  Medical  Informatics,  AMC-­‐UvA   a.c.ravelli@amc.uva.nl  

 

Location  of  Scientific  Research  Project  

Department  of  Medical  Informatics/Department  of  Obstetrics  and  Gynecology   AMC-­‐UvA  

Meibergdreef  15   1105  AZ  Amsterdam    

Practice  teaching  period   October  2014-­‐April  2015    

 

 

 

 

 

 

 

(3)

Acknowledgments  

 

I  would  like  to  express  my  profound  gratitude  to  all  those  who  made  it  possible  for  me   to   complete   this   thesis   project   successfully.   My   deepest   and   sincerest   thanks   to   my   mentor   Sabine   Ensing   without   whose   supervision,   guidance   and   input   none   of   this   would   have   been   possible.   From   the   inception   till   the   completion   of   this   research   project  Sabine  helped  me  to  clarify  my  research  objectives,  she  was  always  available  to   provide  help  or  feedback  when  I  needed  it  and  her  warm  and  encouraging  demeanor   helped   me   maintain   a   positive   attitude   and   work   diligently   throughout   the   entire   process.    

 

I  would  also  like  to  thank  Anita  Ravelli  and  Ameen  Abu-­‐Hanna  who  were  available  to   provide  feedback  and  support  when  I  needed  it.  

 

The  development  of  the  clinical  prediction  tool  would  not  have  been  possible  without   the  input  of  Ewoud  Schuit,  who  conceived  the  idea  for  the  project  a  while  ago  and  also   provided  invaluable  assistance  while  writing  the  code  to  develop  the  tool.      

 

I   am   also   grateful   to   the   staff   and   support   system   of   the   Medical   Informatics   department  at  the  AMC  and  to  my  fellow  graduate  students.  All  the  tips,  advice  and   suggestions  that  I  have  received  during  my  time  at  the  AMC  helped  me  to  navigate  the   Medical  Informatics  master’s  program  successfully.  

 

Lastly,  I  could  never  thank  my  family  enough  for  all  their  love  and  support,  which  has   sustained   me   throughout   my   time   in   Amsterdam-­‐   Leo,   Ola,   Okechukwu,   Ndiya   and   Kachi,  I  am  the  luckiest  girl  in  the  world  to  be  part  of  such  an  amazing  and  supportive   family.  And  also  to  my  best  friends  Cindy  and  Hodan,  you  mean  the  world  to  me,  thank   you  for  everything.    To  the  rest  of  my  friends  and  family  in  Amsterdam  and  abroad,   thank  you  for  always  being  there  for  me  and  for  believing  in  me.  

   

 

 

 

 

 

 

 

 

 

 

 

 

 

Contents  

(4)

Summary                                                                                                                                                                                                                                                                                  5    

Samenvatting                                                                                                                                                                                                                                                                  6    

1        Introduction                                                                                                                                                                                                                                                          7  

1.1 Objective  and  background  information……….7  

1.2 Outline  of  this  thesis………...8  

Bibliography………9  

  2 Prognosis  and  Prognostic  Models  in  health  care                                                                                                                        10   2.1 Introduction………...10  

2.2 Prognosis  and  Prognostic  models………...10  

2.3 Development  of  prognostic  models………...…11  

2.4 Performance  of  prognostic  models:  discrimination  and  calibration………12  

2.5 Internal  and  external  validation  of  prognostic  models……….12  

Bibliography………...………...13  

  3 Labor  and  its  potential  adverse  outcomes;  fetal  asphyxia  and  failure  to  progress    14   3.1 Introduction………...14   3.2 Methods………...15   3.3 Results……….15   3.4 Discussion………...…………..20   3.5 Bibliography………..………….20     4 Evaluation   of   the   existing   prognostic   models   for   fetal   asphyxia   and   failure   to   progress                                                                                                                                                                                                                                                                    23   4.1 Introduction………..…….23   4.2 Methods……….…..24   4.3 Results……….24   4.4 Discussion……….27   Bibliography……….…………28     5.  Development  of  a  clinical  prediction  tool  to  support  clinicians  in  the  assessment  of   the  risks  of  fetal  asphyxia  and  failure  to  progress  in  term  pregnancies                                                        30   5.1  Introduction………..….30   5.2.  Methods……….…….31   5.3  Results……….34   5.4  Discussion  ………38   5.5  Conclusion………42   Bibliography……….……42    

6  Discussion,  conclusions  and  future  recommendations                                                                                                          45    

List  of  Abbreviations                                                                                                                                                                                                                                      43    

Appendices  

(5)

 

The  development  of  a  clinical  prediction  tool  to  support  clinicians  in  the  assessment  of   the  risks  of  fetal  asphyxia  and  failure  to  progress  in  term  pregnancies.  

 

Objective  

During  labor,  problems  can  occur  in  the  child  (fetal  asphyxia)  or  in  the  mother  (failure   to  progress).  Both  problems  can  occur  simultaneously.  However,  at  present  clinicians   base  their  clinical  decisions  typically  on  one  of  these  problems,  rather  than  integrating   both  dimensions.  The  aim  of  this  study  was  to  develop  a  clinical  prediction  tool  that   simultaneously  assesses  the  risk  of  failure  to  progress  (FTP)  and  fetal  asphyxia  and  can   be  applied  in  both  primary  and  secondary  obstetric  care  settings.  

 

Methods  

To  develop  the  prediction  models  for  fetal  asphyxia  as  and  failure  to  progress,  data  on   term   singleton   pregnancies   from   the   Perinatal   Registry   of   the   Netherlands   (PRN)   between   2000   to   2010   were   used.   Bootstrapping   techniques   were   used   for   internal   validation.   Discrimination   (AUC)   and   calibration   (graph,   c-­‐statistics)   were   used   to   assess  the  predictive  performance  of  both  models.    

 

Results  

Two  prediction  models:  one  for  fetal  asphyxia  and  one  for  obstetric  intervention  due   to  failure  to  progress  were  developed.  In  a  summary  graph,  the  predicted  probabilities   of  fetal  asphyxia  versus  the  predicted  probabilities  of  an  intervention  due  to  FTP  and   the  number  of  women  within  each  10th  percentile  combination  of  both  outcomes  were   shown.  The  probability  of  fetal  asphyxia  varied  between  0.1%  and  13%,  whereas  the   probability  of  an  obstetric  intervention  due  to  failure  to  progress  varied  between  0.3%   and   100%.   Overall,   the   chance   of   an   intervention   due   to   FTP   increased   with   an   increased  risk  of  fetal  asphyxia.  However,  in  some  cases  the  risk  of  FTP  is  high  while   the  risk  of  fetal  asphyxia  is  low  and  vice  versa.  To  aid  clinicians  in  the  use  of  the  tool  we   suggested   a   list   of   eight   interventions   (in   increasing   order   of   the   number   of   women   within   each   10th   percentile   combination)   for   various   combinations   of   a   range   of   predicted  probabilities  for  both  outcomes.    

 

Conclusion  

In  women  with  a  singleton  term  pregnancy  in  cephalic  presentation  a  graph  combining   the   risks   of   fetal   asphyxia   and   an   intervention   due   to   FTP   could   aid   clinicians   in   the   choice   of   interventions   during   labor   and   delivery   in   both   primary   and   secondary   obstetric  care  settings.  

 

Key  words:  fetal  asphyxia;  failure  to  progress;  prediction  models               Samenvatting  

(6)

 

De   ontwikkeling   van   een   klinisch   predictie   instrument   om   clinici   te   ondersteunen   bij   het  bepalen  van  de  risico’s  op  feutale  asphyxie  en  stagnatie  van  de  bevalling.  

  Doel  

Tijdens  de  bevalling  kunnen  verschillende  problemen  optreden  zoals  featale    asphyxie   of   stagnatie   van   de   bevalling.   Beide   problemen   kunnen   tegelijkertijd   voorkomen.   Echter  clinici  baseren  hun  klinische  beslissing  vaak  op  een  van  deze  twee  problemen  in   plaats  van  beide  dimensies  te  integreren.    Het  doel  van  deze  studie  is  het  ontwikkelen   van   een   klinisch   predictie   instrument   dat   zowel   het   risico   op   feutale   asphyxie   als   stagnatie  van  de  bevalling  voorspelt.    

 

Methode  

Met  gebruik  van  data  uit  de  Nederlandse  prenatale  registers    (PRN)  hebben  we  twee   predictie   modellen   ontwikkelt   om   feutale   asphyxie   en   stagnatie   van   de   bevalling   te   voorspellen.   Op   basis   van   deze   twee   predictie   modellen   hebben   we   een   tweedimensionaal   predictie   instrument   ontwikkelt   waar   voor   iedere   bevalling   zowel     de  kans  op  feutale  asphyxie  als  stagnatie  van  de  bevalling  weergegeven  wordt  op  een   tweedimensionale  schaal.  

 

Resultaten  

We  lieten  de  predictieve  kans  op  feutale  asphyxie  versus  de  predictieve  kans  van  een   interventie   door   stagnatie   van   de   bevalling   en   het   aantal   vrouwen   binnen   de   combinatie   van   beide   uitkomsten   van   elk   10de   percentiel,   zien.   De   kans   op   feutale   asphyxie   varieerde   tussen   de   0.1%   en   13%,   waar   de   kans   op   een   obstetrische   interventie  door  stagnatie  van  de  bevalling  varieerde  tussen  de  0.3%  en  100%.  De  kans   op  een  interventie  door  stagnatie  van  de  bevalling  nam  toe  bij  een  toenemend  risico   van  feutale  asphyxie.  Echter,  in  sommige  gevallen  is  het  risico  van  stagnatie  tijdens  de   bevalling  hoog  terwijl  het  risico  op  feutale  asphyxie  laag  is  en  vice  versa.  Om  de  clinici   te   ondersteunen   in   het   gebruik   van   het   instrument   hebben   we   een   lijst   van   acht   interventies   opgesteld   (in   toenemende   volgorde   van   het   aantal   vrouwen   binnen   de   combinatie   van   elk   10de   percentiel)   voor   verschillende   combinaties   van   een   reeks   predictieve  kansen  voor  beide  uitkomsten.  

 

Conclusie  

Bij  vrouwen  met  eenling  zwangerschap  die  zich  presenteren  in  een  eerste  of  tweede   lijn   obstetrische   (verloskundige   instelling,   obstetrie)   zorginstelling,   kan   een   grafische   weergave  van  de  risico’s  op  feutale  asphyxie  en  het  inzetten  van  een  interventie  door   het  stagneren  van  de  bevalling  als  bruikbaar  instrument  worden  ingezet  door  de  clinici   om   deze   te   begeleiden   in   de   keuze   voor   het   inzetten   van   een   interventie   tijdens   de   bevalling.  

               

Kernwoorden:  feutale  asphyxie;  stagnatie  van  de  bevalling;  predictie  modellen              

   

(7)

 

Introduction    

The   Scientific   Research   Project   (SRP)   is   a   mandatory   part   of   the   Master   program   of   Medical  Informatics,  and  results  in  a  Master  thesis.  Its  goal  is  to  develop  the  student’s   scientific   problem-­‐focused   approach   and   to   improve   their   ability   to   pursue   lifelong   learning.  Students  will  train  in  undertaking  critical  assessment  of  scientific  biomedical   and   biomedical   informatics   literature,   formulating   clear   research   questions,   and   independently  resolving  information  problems  in  biomedicine  and  reporting  on  it.            This  thesis  describes  the  SRP  called  ‘The  development  of  a  clinical  prediction  tool  to   support  clinicians  in  the  assessment  of  fetal  asphyxia  and  failure  to  progress  in  term   women”,  performed  at  the  Department  of  Medical  Informatics  in  conjunction  with  the   Department  of  Obstetrics  and  Gynecology  at  the  Amsterdam  Medical  Center  (AMC).  In   this  section,  first  we  will  describe  the  goal  of  this  SRP  and  provide  some  background   information,  next  we  will  state  the  research  questions  to  be  answered  by  this  SRP  and   finally  we  will  provide  the  outline  for  this  thesis.  

 

1.1 Objective  and  background  information    

Prognosis   is   defined   as   the   prediction   of   the   future   course   or   outcome   of   disease   processes  [1,2].  For  example  it  might  be  important  to  raise  the  question:  what  are  the   risks   of   maternal   morbidity   and   fetal   mortality   during   labor?   Because   of   various   sources  of  uncertainty  it  is  usually  more  useful  to  estimate  the  probability  of  an  event,   such  as  mortality,  than  to  predict  the  event  itself.  Prognosis  is  required,  among  other   options,   for   making   informed   treatment   decisions,   and   clinicians   are   constantly   prognosticating  events  although  they  might  not  always  be  conscious  of  this  fact.  While   some  clinicians  rely  on  intuition  to  arrive  at  a  prognosis,  others  rely  on  some  formal  or   semi-­‐formal  instrument  such  as  a  risk-­‐calculator.  Some  of  the  approaches  for  outcome   prediction  are  subjective,  relying  on  the  clinicians’  assessments,  while  others  rely  on   models  based  on  collected  data.    

           In  obstetric  care,  the  caregiver  constantly  needs  to  weight  the  risk  of  an  adverse   pregnancy   outcome   i.e.   neonatal   and   maternal   morbidity   or   even   mortality,   when   deciding  between  continuing  labor  versus  obstetric  interventions.  Two  main  adverse   outcomes   that   can   occur   are   failure   of   labor   to   progress   (in   the   woman)   and   fetal   asphyxia   (in   the   child).   Both   outcomes   can   occur   simultaneously.   In   both   cases,   an   intervention  like  a  cesarean  section  or  an  assisted  vaginal  delivery  might  be  necessary.   Recently,  two  prognostic  models  were  developed  to  predict  the  risk  of  fetal  asphyxia   [3]  as  well  as  non-­‐progressive  labor  [4,5].  These  prediction  models  were  developed  on   Dutch   cohorts   of   high-­‐risk   pregnancies   and   the   information   gleaned   from   their   analyses  will  serve  as  the  basis  for  this  research  project.  

         During  this  traineeship  we  will  develop  two  new  models  to  predict  the  risk  of  fetal   asphyxia  as  well  as  non-­‐progressive  labor,  based  on  a  thorough  understanding  of  the   prediction   models   that   have   already   been   developed   and   also   by   including   low-­‐risk   populations   and   incorporating   new   registered   variables   we   think   may   also   be   candidate  predictive  factors  for  both  outcomes.  The  aim  of  this  traineeship  will  be  to   develop   a   two-­‐dimensional   prediction   tool,   based   on   the   new   prediction   models,   in   which  for  each  laboring  woman  the  two  predictions  will  be  expressed  as  coordinates  

(8)

on   a   two-­‐dimensional   plane   e.g.   3%   risk   of   fetal   asphyxia   and   20%   risk   of   non-­‐ progressive  labor.  As  a  complement  to  aid  clinicians  in  the  use  of  the  tool,  we  will  also   suggest   a   list   of   possible   interventions   that   may   be   employed,   depending   on   the   combination   of   predictions   derived   for   each   individual   woman   to   whom   the   tool   is   applied  during  labor.  

 

1.2  Outline  of  this  thesis:  description  of  the  studies  and  research  questions    

The   first   part   of   the   study   presented   in   chapter   two   of   this   thesis   provides   some   background   information   of   prognosis   and   prognostic   models   in   health   care.   The   research  questions  posed  in  this  section  of  the  study  are:  

• How  are  prognostic  models  developed  and  evaluated?  

         The   second   part   of   the   study   presented   in   chapter   three   focuses   on   the   adverse   outcomes  of  labor  that  are  of  particular  interest  in  this  study:  fetal  asphyxia  and  failure   to  progress.  The  research  questions  posed  in  this  section  of  the  study  are:  

• What   is   fetal   asphyxia?   How   is   it   defined,   diagnosed   and   what   are   the   risk   factors  associated  with  fetal  asphyxia?  

• What  is  failure  to  progress?  How  is  it  defined,  diagnosed  and  what  are  the  risk   factors  associated  with  non-­‐progressive  labor?  

         The  third  part  of  the  study  presented  in  chapter  four  of  this  thesis  is  a  summary  of   the   existing   prognostic   models   for   fetal   asphyxia   and   non-­‐progressive   labor.     Knowledge   of   the   development   of   these   prognostic   models   will   guide   us   in   the   development  of  new  prediction  models  for  fetal  asphyxia  and  failure  to  progress  and   subsequently   aid   in   the   development   of   the   clinical   prediction   tool,   which   will   be   discussed  in  the  fourth  part  of  this  thesis.  The  research  questions  posed  in  this  section   of  the  study  are:  

• What  are  the  existing  prognostic  models  for  fetal  asphyxia  and  non-­‐progressive   labor?    

• How  were  these  models  developed  and  evaluated?  How  well  do  these  models   perform  and  what  are  their  limitations?  

       The   fourth   part   of   the   study   presented   in   chapter   five   of   this   thesis   is   the   development   of   a   new   prognostic   models   to   predict   the   risk   of   fetal   asphyxia   and   failure  to  progress  using  a  cohort  of  high  and  low  risk  singleton,  term  pregnancies.  By   combining  these  models  we  developed  a  two-­‐dimensional  prediction  tool,  in  which  for   each   laboring   woman   the   two   predictions   were   expressed   as   coordinates   on   a   two-­‐ dimensional  plane.    The  research  questions  posed  in  this  study  are:    

• Which   combination   of   predictive   factors   in   the   prognostic   model   for   fetal   asphyxia  performs  best?    

• Which  combination  of  predictive  factors  in  the  prognostic  model  for  failure  to   progress  performs  best?  

• How  can  the  two-­‐dimensional  clinical  decision  tool  be  used  to  guide  clinicians   in  the  choice  of  interventions  during  labor?  

         The   final   part   of   the   study,   which   will   be   presented   in   chapter   six,   provides   the   discussion  and  conclusions  we  arrived  at  during  the  study  as  well  as  recommendations   to  develop  the  clinical  tool  further.  

 

(9)

 

1. A  Abu-­‐Hanna  and  PJF  Lucas.  Prognostic  models  in  medicine:  AI  and  statistical   approaches.  Method  Inform  Med,  40:1_5,  2001.  

2. PJF  Lucas  and  A  Abu-­‐Hanna.  Prognostic  methods  in  medicine.  Artif  Intell  Med,   15:105_119,  1999.  

3. Michelle  E.M.H.  Westerhuis  et  al.  Prediction  of  Neonatal  Metabolic  Acidosis  in   Women  with  a  Singleton  Term  Pregnancy  in  Cephalic  Presentation.  American   Journal  of  Perinatology  2012;  29:167-­‐174.  

4. E  Schuit  et  al.  A  clinical  prediction  model  to  assess  the  risk  of  operative   delivery.  BJOG  2012;  119:915-­‐923  

5. S.  Katherine  Laughton  et  al.  Using  a  Simplified  Bishop  Score  to  Predict  Vaginal   Delivery.  Obstet  Gynecol  2011;  117:805-­‐11  

                                                                    Chapter  2  

(10)

 

Prognosis  and  Prognostic  models  in  health  care    

2.1  Introduction    

In  this  chapter  background  information  on  the  domain  will  be  given,  beginning  with  an   explanation  of  the  term  'prognosis'  and  the  meaning  of  the  term  in  healthcare  [1,2].   This   will   be   followed   by   an   explanation   of   the   development   and   the   evaluation   and   validation  of  prognostic  models.  

 

2.2  What  is  prognosis?    

Prognosis  simply  means  foreseeing,  predicting  or  estimating  the  probability  or  risk  of   future  conditions.  In  healthcare,  prognosis  commonly  relates  to  the  probability  or  risk   of  an  individual  developing  a  particular  state  of  health  (or  an  outcome)  over  a  specific   period  of  time,  based  on  his  or  her  clinical  and  non-­‐clinical  profile.  Outcomes  may  be   specific   events,   such   as   death   or   complications,   or   quantities,   such   as   disease   progression,  changes  in  pain  or  quality  of  life.  

         In   practice,   clinicians   do   not   predict   the   course   of   an   illness   but   the   course   of   an   illness   in   a   particular   individual.   Prognosis   may   be   shaped   by   a   patient’s   age,   sex,   history,  symptoms,  signs  and  other  test  results.  Besides,  prognostication  in  healthcare   is  not  limited  to  those  who  are  ill.  Healthcare  professionals  regularly  predict  the  future   of   healthy   individuals   e.g.   using   the   Apgar   score   to   determine   the   prognosis   of   newborns   and   prenatal   testing   to   determine   the   risk   that   a   pregnant   woman   will   deliver  a  baby  with  Down’s  syndrome.  

         Given   the   variability   among   patients   and   in   the   etiology,   presentation,   and   treatment   of   diseases   and   other   health   states,   a   single   predictor   or   variable   rarely   gives   an   adequate   estimate   of   prognosis.   Clinicians-­‐implicitly   and   explicitly-­‐   use   multiple   predictors   to   estimate   a   patient’s   prognosis.   Prognostic   studies   therefore   need   to   use   a   multivariable   approach   in   design   and   analysis   to   determine   the   important  predictors  of  the  studied  outcome  probabilities  for  different  combinations   of   predictors,   or   to   provide   tools   to   estimate   such   probabilities.   These   tools   are   commonly  called  prognostic  models,  prediction  models,  prediction  rules,  or  risk  scores.   They   enable   care   providers   to   use   combinations   of   predictor   values   to   estimate   an   absolute  risk  or  probability  that  an  outcome  will  occur  in  an  individual.  A  multivariable   approach  also  enables  researchers  to  investigate  whether  specific  prognostic  factors  or   markers   that   are,   say,   more   invasive   or   costly   to   measure,   have   worthwhile   added   predictive  value  beyond  cheap  or  simply  obtained  predictors.    

         Prognostic   models   are   used   in   various   settings   and   for   various   reasons.   The   main   reasons  are  to  inform  individuals  about  the  future  course  of  their  illness  (or  their  risk   of  developing  illness)  and  to  guide  doctors  and  patients  in  joint  decisions  on  further   treatment.   An   example   from   obstetric   care   would   include   the   use   of   a   simplified   Bishop  score  to  predict  the  chances  of  vaginal  delivery  [1,2].  

       

(11)

2.3  Development  of  prognostic  models    

In   developing   a   prognostic   model   there   are   certain   non-­‐statistical   and   statistical   aspects  that  should  be  taken  into  account  [3].  The  non-­‐statistical  characteristics  of  the   multivariable  study  aimed  at  developing  a  prognostic  model  include:  

 

Objective  

The   main   objective   of   a   prognostic   study   is   to   determine   the   probability   of   the   specified   outcome   with   different   combinations   of   predictors   in   a   well-­‐defined   population.  

 

Study  sample  

The   study   sample   includes   people   at   risk   of   developing   the   outcome   of   interest,   defined  by  the  presence  of  a  particular  condition.  

 

Study  design  

The  best  design  to  answer  prognostic  questions  is  a  cohort  study.  A  prospective  study   is  preferable  as  it  enables  optimal  measurement  of  predictors  and  outcome.  

 

Predictors  

Candidate   predictors   can   be   obtained   from   patient   demographics,   clinical   history,   physical   examination,   disease   characteristics,   test   results   and   previous   treatments.   Studied   predictors   should   be   clearly   defined,   standardized   and   reproducible   to   enhance   generalizability   and   application   of   the   study   results   to   practice.   Predictors   should   be   measured   using   methods   applicable   to   daily   practice.   Finally,   prognostic   studies   should   only   include   predictors   that   will   be   available   at   the   time   when   the   model  is  intended  to  be  used  i.e.  if  the  aim  is  to  predict  a  patient’s  prognosis  at  the   time  of  diagnosis,  predictors  that  will  not  be  known  until  actual  treatment  has  started   are  of  little  value.  

 

Outcome  

Prognostic   studies   should   focus   on   outcomes   that   are   relevant   to   patients   such   as   death,  complications,  treatment  response  or  quality  of  life.  The  period  over  which  the   outcome   is   studied   and   the   methods   or   measurement   should   be   clearly   defined.   Finally,   outcomes   should   be   measured   without   knowledge   of   the   predictors   under   study  to  prevent  bias  [2].  

 

Some   of   the   important   statistical   aspects   to   consider   when   developing   a   prognostic   model  include:  

 

Selecting  candidate  predictors  

Studies   often   measure   more   predictors   than   can   sensibly   be   used   in   a   model   and   selection   is   required.   Predictors   already   reported   as   prognostic   would   normally   be   included.   Predictors   that   are   highly   correlated   with   others   contribute   little   independent  information  and  may  be  excluded  beforehand.  However,  predictors  that   are  not  significant  in  a  univariable  analysis  should  not  be  excluded  as  candidates.    

(12)

 

Evaluating  data  quality  

In  principle,  data  used  for  developing  a  prognostic  model  should  be  fit  for  the  purpose.   Measurements   of   candidate   predictors   and   outcomes   should   be   comparable   across   clinicians  or  study  centers.  Modern  statistical  techniques  such  as  multiple  imputations   can  handle  data  sets  with  missing  values.  

 

Selecting  variables  

No   agreement   exists   on   the   best   method   for   selecting   variables.     In   the   full   model   approach  all  the  candidate  variables  are  included  in  the  model.  This  model  is  claimed   to   avoid   over   fitting   and   selection   bias   and   provide   correct   standard   errors   and   p   values.  The  backward  elimination  approach  starts  with  all  the  candidate  variables.  A   nominal   significance   level,   often   5%   is   chosen   in   advance.   A   sequence   of   hypothesis   tests   is   applied   to   determine   whether   a   given   variable   should   be   removed   from   the   model.  Backward  elimination  is  preferable  to  forward  selection  whereby  the  model  is   built  up  from  the  best  candidate  predictors  [3].    

 

2.4  Performance  of  prognostic  models:  calibration  and  discrimination          

Prognostic  models  yield  scores  to  enable  the  prediction  of  the  risk  of  future  events  in   individual  patients  or  groups  and  the  stratification  of  patients  by  these  risks.  A  good   model  may  allow  the  reasonably  reliable  classification  of  patients  into  risk  groups  with   different   prognoses.   However,   to   show   that   a   prognostic   model   is   valuable,   it   is   not   sufficient  to  show  that  it  successfully  predicts  outcome  in  the  initial  development  data.   We  need  evidence  that  the  model  performs  well  for  other  groups  of  patients.    

         The   performance   of   a   logistic   regression   model   may   be   assessed   in   terms   of   calibration  and  discrimination.    

         Calibration   can   be   investigated   by   plotting   the   observed   proportions   of   events   against  the  predicted  risks  for  groups  defined  by  ranges  of  individual  predicted  risks.   Ideally,  if  the  observed  proportion  of  events  and  predicted  probabilities  agree  over  the   whole  range  of  probabilities,  the  plot  shows  a  45°  line.  This  plot  can  be  accompanied   by  the  Hosmer-­‐Lemeshow  test.  The  overall  observed  and  predicted  event  probabilities   are  by  definition  equal  for  the  sample  used  to  develop  the  model.    

         Various   statistics   can   summarize   discrimination   between   individuals   with   and   without   the   outcome   event.   The   area   under   the   receiver   operating   curve   or   the   equivalent  c  (concordance)  index  is  the  chance  that  given  two  patients,  one  who  will   develop  an  event  and  the  other  will  not,  the  model  will  assign  a  higher  probability  of   an  event  to  the  former.    

 

2.5  Internal  and  external  validation  of  prognostic  models    

A  model’s  validity  can  be  assessed  using  new  recent  data  from  the  same  source  as  the   derivation   sample   (“apparent   validation”),   but   a   true   external   validation   of   the   prediction  model’s  generalizability  requires  evaluation  on  data  from  elsewhere  [4].      

   

(13)

Internal  validation    

One  validation  strategy  is  internal  validation.  One  approach  to  internal  validation  is  to   randomly  split  the  dataset  into  two  parts  and  the  model  is  developed  using  the  first   portion   (often   called   the   training   set)   and   its   predictive   accuracy   is   assessed   on   the   second  portion.      

         Another  validation  approach  is  non-­‐random  splitting  which  may  be  preferable  as  it   reduces  the  similarity  of  the  two  sets  of  patients.  If  the  available  data  are  limited,  the   model  can  be  developed  on  the  whole  dataset  and  techniques  of  data  re-­‐use,  such  as   cross-­‐validation  and  bootstrapping  can  be  used  to  assess  performance.  Bootstrapping   is   a   method   of   estimating   properties   of   an   estimator,   such   as   its   variance,   by   measuring   those   properties   when   sampling   from   an   approximate   distribution   e.g.   a   resample  with  replacement,  of  the  observed  dataset  [4].    

 

Quality  assessment    

There  are  a  few  frameworks  that  have  been  developed  for  the  purpose  of  assessing   the  quality  of  studies  describing  the  development  or  validation  of  prediction  models   [5,6].    One  such  framework  is  the  Transparent  Reporting  of  a  multivariable  prediction   model  for  individual  prognosis  or  diagnosis  (TRIPOD)  statement  [7].    

 

Bibliography    

1. A  Abu-­‐Hanna  and  PJF  Lucas.  Prognostic  models  in  medicine:  AI  and   statistical  approaches.  Method  Inform  Med,  40:1_5,  2001  

2. Karel  GM  Moons  et  al.  Prognosis  and  prognostic  research:  what,  why  and  how?   BMJ  2009;338:b375  

3. Karel  GM  Moons  et  al.  Prognosis  and  prognostic  research:  Developing  a   prognostic  model.  BMJ  2009;338:b604  

4. Douglas  G  Altman  et  al.  Prognosis  and  prognostic  research:  validating  a   prognostic  model.  BMJ  2009;  338:b605  

5. Medlock  S  et  al.  Prediction  of  mortality  in  very  premature  infants:  A  systematic   review  of  prediction  models.  PLoS  ONE  2011;  6(9)  e:  23441  

6. Leushuis  E  et  al.  Prediction  models  in  reproductive  medicine:  a  critical   appraisal.  Hum.  Reprod.  Update  (2009)  15  (5):  537-­‐552.  

7. Collins  SG  at  el.  Transparent  reporting  of  a  multivariable  prediction  model  for   individual  prognosis  or  diagnosis  (TRIPOD):  the  TRIPOD  statement.  BMJ  2015;   350:g7594                   Chapter  3  

(14)

 

Labor  and  its  potential  adverse  outcomes;  fetal  asphyxia  and  failure  to  progress    

3.1  Introduction              

Labor  is  the  process  by  which  the  fetus  and  the  placenta  leave  the  uterus.  Delivery  can   occur   in   two   ways:   vaginally   or   by   a   cesarean   section.   According   to   the   American   Pregnancy   Association,   labor   usually   occurs   in   three   stages   [1].   The   first   stage   is   defined   as   the   time   from   the   onset   of   labor   until   the   cervix   is   completely   dilated   to   10cm.  The  second  stage  of  labor  is  defined  as  the  period  after  the  cervix  is  dilated  to   10cm  until  the  baby  is  delivered.  The  third  stage  of  labor  involves  the  delivery  of  the   placenta.  The  first  stage  of  labor  is  the  longest  and  involves  three  phases:  early,  active   and  transition  labor  phase.    The  early  labor  phase  is  the  time  from  the  onset  of  labor   until   the   cervix   is   dilated   to   about   3cm.   The   active   labor   phase   continues   from   3cm   until  the  cervix  is  dilated  to  7cm.  The  transition  phase  continues  from  7cm  until  the   cervix  is  fully  dilated  to  10cm.  

           For  laboring  women  the  preferred  pathway  and  outcome  would  be  a  spontaneous   labor  resulting  in  a  vaginal  delivery.  However,  in  some  cases  labor  can  take  different   pathways   and   involve   different   interventions,   some   of   which   may   result   in   adverse   outcomes  for  the  mother  and  child.  For  instance,  induction  of  labor  is  a  common  and   essential   element   of   obstetric   practice   with   an   incidence   of   approximately   20%   of   pregnancies  [2,3].  If  induction  is  successful  the  laboring  woman  will  achieve  the  active   phase  of  labor.    Some  women  may  also  experience  a  failure  of  labor  to  progress;  in   first  stage  of  labor  during  the  active  phase  this  implies  no  cervical  dilation  for  at  least  2   hours  and  in  the  second  stage  of  labor  this  implies  no  descent  of  the  fetus’s  head  for  at   least  one  hour  despite  adequate  uterine  contractions.  If  fetal  distress  is  suspected  or   failure   to   progress   is   observed   during   labor,   the   clinician   might   choose   to   intervene   through  an  assisted  vaginal  delivery  or  by  performing  a  cesarean  section.    

           Each  of  these  interventions:  induction  of  labor,  instrumental  delivery  and  cesarean   section  should  be  used  with  caution  as  they  may  increase  the  risk  of  adverse  outcome   for  either  the  mother  or  the  baby.  For  example,  induction  of  labor  has  been  associated   with  increased  rates  of  epidural  anesthesia,  emergency  cesarean  delivery  and  adverse   neonatal   events   such   as   requirement   for   resuscitation   [4,5].   Similarly,   instrumental   delivery  is  a  strong  risk  factor  third-­‐  and  fourth-­‐degree  perineal  injuries  [6].  Cesarean   delivery   itself   is   associated   with   an   increased   risk   of   respiratory   morbidity   in   babies,   even  after  37  weeks’  gestation  [7].  

           For  the  purpose  of  this  SRP,  the  outcomes  of  interest  are  fetal  asphyxia  and  failure   of  labor  to  progress  [8].  The  research  questions  posed  in  this  chapter  are:  

• What   is   fetal   asphyxia?   How   is   it   defined,   diagnosed   and   what   are   the   predictive  risk  factors  for  fetal  asphyxia?  

• What   is   failure   to   progress?   How   is   it   defined,   diagnosed   and   what   are   the   interventions  and  predictive  risk  factors  for  non-­‐progressive  labor?  

       

(15)

 

We   searched   PubMed   from   January   1,   1990   until   October   1,   2014   for   publications   studying   fetal   asphyxia   (also   referred   to   in   some   instances   as   intrapartum   asphyxia,   birth  asphyxia  or  metabolic  acidosis)  and  failure  to  progress  (also  referred  to  in  some   instances   as   non-­‐progressive   labor   or   dystocia)   and   offering   a   definition,   means   of   diagnosis  and  classification.  We  also  searched  for  publications  studying  the  risk  factors   (maternal  and  obstetric  risk  factors)  that  are  associated  with  fetal  asphyxia  and  failure   to  progress.  

           The  following  keywords  were  used  in  the  database  search:  fetal  asphyxia,  failure  to   progress,   risk   factors,   cesarean   section,   instrumental   vaginal   delivery   and   labor/delivery.  The  search  was  narrowed  down  to  articles  written  in  English.    

         All  the  titles  and  abstracts  of  the  articles  were  reviewed  and  articles  were  included   if   they   described   a   descriptive   study   of   fetal   asphyxia   or   failure   to   progress   and/or   their  associated  risk  factors.  The  full  text  articles  of  these  studies  were  retrieved  and   all  of  them  were  reviewed.  

         First   we   looked   at   the   definition   of   asphyxia   in   general   and   fetal   asphyxia   in   particular  and  also  the  definition  of  failure  to  progress.  Next  we  looked  at  the  methods   used   to   diagnose   fetal   asphyxia   and   failure   to   progress.   Finally   we   looked   at   the   maternal  and  obstetric  risk  factors  (both  antepartum  and  intrapartum  characteristics)   that   are   associated   with   (and   possibly   predictive   of)   fetal   asphyxia   and   failure   to   progress.    

 

3.3  Results      

After   full   text   reviews,   19   articles   were   found   eligible   for   inclusion   in   the   study.     Of   these,  8  articles  focused  primarily  on  the  definition  and  diagnosis  of  fetal  asphyxia  and   4   articles   focused   primarily   on   the   risk   factors   associated   with   fetal   asphyxia.   Three   articles  focused  primarily  on  the  definition  and  diagnosis  of  failure  to  progress  and  4   articles   focused   primarily   on   the   risk   factors   and   obstetric   interventions   associated   with  failure  to  progress  

 

3.3.1  Fetal  asphyxia    

Definition  of  fetal  asphyxia    

Specific  evidence  of  asphyxia  can  be  provided  by  means  of  a  blood  gas  and  acid-­‐base   assessment.   When   this   has   been   done,   the   timing   of   the   asphyxia   event   can   be   determined   more   accurately.   For   this   reason,   the   use   of   the   term   perinatal   or   birth   asphyxia   is   discouraged.   The   effects   of   fetal   asphyxia   resulting   from   compromised   maternal-­‐fetal   blood   gas   exchange   before   delivery   should   be   differentiated   from   newborn   asphyxia   resulting   from   cardiorespiratory   complications   after   delivery.   Accurate  diagnosis  and  precise  timing  are  important  if  strategies  to  prevent  or  modify   the  effects  of  asphyxia  are  to  be  developed.  

           A  task  force  set  up  by  the  World  Federation  of  Neurology  Group  for  the  prevention   of  cerebral  palsy  and  related  neurologic  disorders  defined  asphyxia  as  a  condition  of   impaired   gas   exchange   leading,   if   it   persists,   to   progressive   hypoxemia,   hypercapnia   and   metabolic   acidosis[9].     By   this   definition   asphyxia   may   affect   the   fetus   and  

(16)

newborn.   In   some   cases,   the   asphyxia   is   short-­‐lived   with   no   apparent   pathological   effects.  Significant  asphyxia  leading  to  tissue  oxygen  debt  with  accumulation  of  fixed   acids  results  in  metabolic  acidosis.  Thus  for  clinical  purposes  the  task  force  proposed   the  following  definition  for  fetal  asphyxia:  a  condition  of  impaired  blood  gas  exchange   leading   to   progressive   hypoxemia   and   hypercapnia   with   a   significant   metabolic   acidosis.    

         A   diagnosis   of   an   asphyxiating   event   should   not   be   made   unless   there   is   some   evidence  of  an  interruption  of  oxygen  supply  or  blood  flow  to  the  fetus.  These  events   can   be   secondary   to   problems   from   the   mother   (e.g.   hypotension,   toxemia,   uterine   rupture),  the  placenta  or  umbilical  cord  (e.g.  abruption,  infection  or  inflammation,  or   umbilical   cord   compression),   or   the   fetus   or   infant   (e.g.   central   nervous   system   depression,  anomalies,  infection)  [10].  The  term  “asphyxia”  should  not  be  used  unless   the  neonate  meets  all  of  the  following  conditions:  umbilical  cord  arterial  pH  less  than   7,  Apgar  score  of  0  to  3  for  longer  than  5  minutes,  neurological  manifestations  (e.g.   seizures,  coma  or  hypotonia),  and  multisystemic  organ  dysfunction  [11]  

           The  Apgar  score  was  developed  by  Dr.  Virginia  Apgar  in  1952  as  an  objective  tool   that   measures   five   signs   of   physiologic   adaptation:   heart   rate,   respiratory   effort,   muscle  tone,  reflex  irritability  and  color.  A  score  is  a  sum  of  the  values  assigned  to  the   infant  at  1  and  5  minutes  of  life,  with  a  score  of  7  or  more  indicating  that  the  baby  is  in   good   to   excellent   condition   [12].   A   retrospective   analysis   of   151,891   neonates   born   over  a  10-­‐year  period  revealed  a  mortality  rate  of  24.4%  for  infants  with  five-­‐minute   Apgar   scores   of   0   to   3   versus   a   mortality   rate   of   0.02%   for   infants   with   five-­‐minute   Apgar  scores  of  7  to  10  suggesting  that  an  Apgar  score  of  3  or  less  at  5  minutes  of  life   does  predict  a  higher  rate  of  mortality  [13].  

 

Diagnosis  of  fetal  asphyxia                

Laboratory   and   clinical   studies   suggest   that   the   threshold   for   a   significant   metabolic   acidosis   is   a   base   deficit   between   12   and   16   mmol/L.   Asphyxia   with   significant   metabolic   acidosis   is   associated   with   seizures   in   the   fetal   lamb.   Clinical   studies   have   also  showed  an  association  between  severe  academia  and  multiorgan  complications  in   the  newborn.  

         Routine   blood   gas   and   acid-­‐base   assessment   of   umbilical   artery   blood   at   delivery   has  demonstrated  an  umbilical  artery  base  deficit  >  12mmol/L  in  2%  and  >  16  mmol/L   in  0.5%  of  the  total  population.  This  assessment  means  that  98%  of  newborns  do  not   have  a  significant  asphyxial  episode  during  labor  and  delivery.  However,  for  the  2%  of   newborns  that  have  been  exposed  to  asphyxia  this  may  affect  their  outcome.  

         Evidence  of  a  significant  metabolic  acidosis  can  establish  that  exposure  to  asphyxia   has   occurred.   This   will   also   suggest   the   degree   of   metabolic   acidosis   at   the   time   of   sampling.   However,   it   does   not   necessarily   indicate   the   severity   of   the   asphyxial   exposure  to  the  fetus.  The  duration  of  the  asphyxia  is  generally  not  known.  Also  the   nature   of   the   exposure   (i.e.   continuous   or   intermittent)   or   whether   the   asphyxia   during  labor  and  delivery  is  the  last  in  a  series  of  insults  is  not  known.  

         The   importance   of   an   asphyxial   exposure   is   influenced   by   the   fetal   response.   The   response  to  asphyxia  is  a  concentration  of  the  fetal  circulation  with  increased  blood   flow  to  the  brain,  heart  and  adrenals.  If  the  hypoxia  is  sustained  fetal  cardiovascular   decompensation   will   occur.     Laboratory   studies   in   fetal   lamb   have   shown   that   fetal  

(17)

cardiovascular  decompensation  results  in  decreased  cerebral  blood  flow  and  cerebral   oxygen   metabolism   and   brain   damage.   These   studies   also   suggest   that   the   fetal   response  is  not  necessarily  proportional  to  the  exposure  [14].    

         For   the   2%   of   newborns   who   have   been   exposed   to   an   asphyxial   event,   a   classification   of   the   severity   of   the   exposure   is   required   to   predict   the   long-­‐term   outcome  in  the  child.  The  severity  of  intrapartum  fetal  asphyxia  can  be  classified  by   determining   the   short-­‐term   outcome   as   expressed   by   newborn   encephalopathy   and   other   newborn   organs   system   complications.   This   alternative   is   acceptable   because   the  duration  of  the  asphyxia  itself  cannot  be  determined  and  the  clinical  measures  of   fetal   cardiovascular   compensation   and   decompensation   are   not   available.   Also,   the   susceptibility  to  the  exposure  may  depend  on  the  different  characteristics  of  the  fetus   i.e.  gestational  age,  small  for  gestational  age  vs.  normal  for  gestational  age.  

         The   clinical   signs   of   newborn   encephalopathy   associated   with   intrapartum   fetal   asphyxia  occur  more  often  on  the  first  day  after  delivery,  with  decreasing  frequency  on   the   second   and   third   day   after   delivery.   Newborn   encephalopathy   was   classified   as   minor  if  it  consisted  of  jitteriness  and  irritability;  moderate  if  it  consisted  of  lethargy  or   abnormal   tone   and   severe   if   it   consisted   of   coma   or   abnormal   tone   and   multiple   seizures.   Cardiovascular   complications   were   classified   as   minor   it   there   was   bradycardia   or   tachycardia   (defined   by   the   95%   confidence   limits   for   heart   rate   for   term   and   preterm   newborns),   moderate   if   there   was   hypertension   or   hypotension   (defined   by   the   95%   confidence   limits   for   blood   pressure   for   term   and   preterm   newborns),   and   severe   if   there   was   abnormal   electrocardiographic   or   echocardiographic   findings.   Respiratory   complications   were   classified   as   minor   if   needing   supplementary   oxygen,   moderate   if   requiring   continuous   positive   airway   pressure   or   transient   ventilation   (<24   hours),   or   severe   if   requiring   mechanical   ventilation   for   >24   hours.   Renal   complications   were   classified   as   minor   if   hematuria   was  observed,  moderate  if  there  was  an  elevation  of  serum  creatinine  (>100  µmol/L)   and  severe  with  clinical  evidence  of  oliguria  (<1  ml/kg/hr)  or  anuria.  The  classification   of  mild,  moderate  and  severe  intrapartum  fetal  asphyxia  is  based  on  the  evidence  that   at  this  time  early-­‐onset  newborn  encephalopathy  is  the  best  single  predictor  of  long-­‐ term  outcome.  The  long-­‐term  outcome  examined  in  most  studies  of  fetal  asphyxia  has   been   severe   handicap.   Luckily,   most   survivors   of   intrapartum   fetal   asphyxia   do   not   have  major  sequelae.  However,  it  is  unclear  whether  there  is  a  range  of  casualty  after   fetal  asphyxia  [15].    

         A  study  by  Ruth  and  Raivio  compared  Apgar  scores,  cord  blood  pH  and  cord  lactate   levels  in  more  than  900  infants  and  looked  at  outcome.  They  found  that  11%  of  the   infants  with  cord  blood  acidosis  had  an  Apgar  score  below  7,  whereas  41%  of  infants   with  an  Apgar  of  less  than  7  had  an  acidosis.  In  the  end,  the  sensitivity  and  positive   predictive  value  of  a  low  pH  for  adverse  outcomes  were  21%  and  8%  respectively.  Cord   blood   lactate   levels   were   at   12%   and   5%   respectively.   The   sensitivity   and   positive   predictive   value   of   the   Apgar   score   values   were   12%   and   19%   respectively.   Thus,   in   situations  where  fetal  blood  sample  values  are  not  readily  available  Apgar  scores  may   be   used   to   obtain   a   comparable   diagnosis   of   incidences   of   fetal   asphyxia   i.e.   an   asphyxial   event   is   assumed   to   have   occurred   if   the   Apgar   score   is   less   than   7   at   5   minutes  of  life  [16].  

   

Referenties

GERELATEERDE DOCUMENTEN

Tijdens het melken wordt de melkafgif- te op gang gehouden doordat de spenen voort- durend gemasseerd worden door de beweging van de tepelvoering.. Dit laatste geldt ook voor

Nu inzichtelijk is gemaakt wat de definieerbare verschillen zijn tussen de zelfstandige zonder personeel en de werknemer, zal in deze paragraaf worden gekeken

In five out of the ten cases which were identified to have weak cross channel strategy this had a negative influence on the company’s chances of survival. For

De gemiddeld hogere score op de dimensie openheid voor ervaringen draagt er aan bij dat internal auditors meer dan gemiddeld op zoek gaan naar nieuwe informatie

A strong positive correlation is found between health and safety and the casino employees’ economic and family domain, social domain, esteem domain, actualisation

discussed in the following chapter. On the subject of housing opportunities policies the focus is on 

This in turn leads to reduced Chinese imports, partly from the US (2010, p. Third, Fair claims that higher import prices for the US lead to a higher price level overall, which has

(…) als nu inderdaad door de uitspraak van deze rechter de lokale afdeling PVV Nijmegen inderdaad denkt ooeeh ik moet toch maar uitkijken wat we over die Marokkanen in