Editorials
Performance-based financing in low- and middle-income countries:
still more questions than answers
Atle Fretheim,a Sophie Witter,b Anne Karin Lindahlc & Ingvar Theo Olsend
Performance-based inancing is promot- questionable data quality. he results If conducting an RCT is not feasible,
ed as a promising strategy for improving on antenatal care attendance were also other robust evaluation designs should be
health service delivery and helping to heterogeneous. considered. One option is an interrupted
reach the Millennium Development hese indings clearly show that no time series, in which outcome data are
Goals.1,2 But what is the evidence sup- general conclusion can be drawn regard- collected at regular intervals during base-
porting its use? ing the likely impact of performance- line and post-intervention periods (e.g.
Performance-based inancing may based inancing in low-and middle- for one year before and ater). his type of
be deined as “the transfer of money income countries. For one thing, most study, however, normally requires access
or material goods conditional on tak- of the studies found through the review to reasonably good routine data from hos-
ing a measurable action or achieving a were methodologically weak and had pital records or other reliable sources not
predetermined performance target”.2 We poor internal validity. Furthermore, always available in low-resource settings.
recently conducted a Cochrane review of since the impact of complex interven- Rather than focus exclusively on
the available evidence on the efective- tions such as performance-based i- targeted indicators, future performance-
ness of performance-based inancing nancing depends largely on the context based inancing evaluations should
in low- and middle- income countries.3 in which they are implemented, results consider untargeted efects and systems
Nine studies, of which fewer than half may vary. Finally, the studies difered efects by starting with a conceptual
had been published in scientiic jour- substantially in the way in which the framework of how the intervention might
nals, fulilled our inclusion criteria: one performance-based inancing scheme work and what the important outcome
randomized controlled trial (RCT) and was designed and implemented. measures are. Also, study protocols and
two interrupted time series conducted In our view, minimizing the risk evaluation plans should be prepared and
in Asia, and six controlled before–ater of producing biased results in future published as early as possible, preferably
studies conducted in Africa. Only two performance-based inancing evalua- before programme implementation. It is
outcomes related to health care uti- tions is essential. Since performance- of concern that, in a search conducted
lization – institutional deliveries and based inancing schemes are not likely to on 3 April 2012, we found no ongoing
antenatal care – were assessed in more have a large impact, efect estimates need performance-based inancing trials from
than one trial. Inconsistent results across to be as reliable and precise as possible low- and middle-income countries in the
studies made summarizing and inter- to avoid missing real efects or being International Clinical Trials Registry Plat-
preting the evidence diicult. misled by seemingly positive indings. form. We hope that more and better evalu-
he most rigorous African study hus, trials should be conducted in set- ations will emerge in future updates of our
reported a moderate increase in insti- tings where a suiciently large number Cochrane review to enable us to formulate
tutional deliveries, from around 35% of facilities or geographical areas can be forthright conclusions about the efects of
to 42% (i.e. 7 percentage points; 95% randomized. performance-based inancing in low- and
conidence interval, CI: 1–14).4 Find- Sometimes RCTs are too expensive, middle-income countries. ■
ings from studies in Burundi, 5 the time-consuming or impractical. he
Democratic Republic of the Congo 6 most recent study of performance-based Competing interests: ITO is employed by
and Rwanda7 showed disparate indings: inancing, from Rwanda, illustrates the the Norwegian Agency for Develop-
one reported a signiicant increase in practical setbacks researchers some- ment Cooperation (Norad), which funds
institutional deliveries, another found times face. he study was set up as a performance-based inancing projects in
little or no change and the third showed well-designed RCT with districts ran- several countries. Also, the Norwegian
a signiicant decrease. For these indings domly assigned to performance-based government is one of the contributors to
the risk of bias is high, partly because inancing intervention and non-inter- he World Bank’s Health Results Innova-
intervention and control areas were not vention groups,4 but right before the tion Trust Fund (HRITF).
randomly allocated and the same people trial began, the government redeined
who implemented the programmes the administrative district boundaries References
also evaluated them. Two additional and several of the districts in the control Available at: http://www.who.int/bulletin/
studies reporting on institutional de- group were shited to the intervention volumes/90/8/12-106468
liveries were programme evaluations group. his may have weakened the
with a substantial risk of bias due to validity of the study indings.
a
Norwegian Knowledge Centre for the Health Services, PO Box 7004, St Olavs plass, Oslo, 0130, Norway.
b
Immpact, University of Aberdeen, Aberdeen, Scotland.
c
Norwegian Knowledge Centre for the Health Services, Oslo, Norway.
d
Norwegian Agency for Development Cooperation, Oslo, Norway.
Correspondence to Atle Fretheim (e-mail: ).
Bull World Health Organ 2012;90:559–559A | doi:10.2471/BLT.12.106468 559
Performance-based financing in low- and middle-income countries:
still more questions than answers
Atle Fretheim,a Sophie Witter,b Anne Karin Lindahlc & Ingvar Theo Olsend
Performance-based inancing is promot- questionable data quality. he results If conducting an RCT is not feasible,
ed as a promising strategy for improving on antenatal care attendance were also other robust evaluation designs should be
health service delivery and helping to heterogeneous. considered. One option is an interrupted
reach the Millennium Development hese indings clearly show that no time series, in which outcome data are
Goals.1,2 But what is the evidence sup- general conclusion can be drawn regard- collected at regular intervals during base-
porting its use? ing the likely impact of performance- line and post-intervention periods (e.g.
Performance-based inancing may based inancing in low-and middle- for one year before and ater). his type of
be deined as “the transfer of money income countries. For one thing, most study, however, normally requires access
or material goods conditional on tak- of the studies found through the review to reasonably good routine data from hos-
ing a measurable action or achieving a were methodologically weak and had pital records or other reliable sources not
predetermined performance target”.2 We poor internal validity. Furthermore, always available in low-resource settings.
recently conducted a Cochrane review of since the impact of complex interven- Rather than focus exclusively on
the available evidence on the efective- tions such as performance-based i- targeted indicators, future performance-
ness of performance-based inancing nancing depends largely on the context based inancing evaluations should
in low- and middle- income countries.3 in which they are implemented, results consider untargeted efects and systems
Nine studies, of which fewer than half may vary. Finally, the studies difered efects by starting with a conceptual
had been published in scientiic jour- substantially in the way in which the framework of how the intervention might
nals, fulilled our inclusion criteria: one performance-based inancing scheme work and what the important outcome
randomized controlled trial (RCT) and was designed and implemented. measures are. Also, study protocols and
two interrupted time series conducted In our view, minimizing the risk evaluation plans should be prepared and
in Asia, and six controlled before–ater of producing biased results in future published as early as possible, preferably
studies conducted in Africa. Only two performance-based inancing evalua- before programme implementation. It is
outcomes related to health care uti- tions is essential. Since performance- of concern that, in a search conducted
lization – institutional deliveries and based inancing schemes are not likely to on 3 April 2012, we found no ongoing
antenatal care – were assessed in more have a large impact, efect estimates need performance-based inancing trials from
than one trial. Inconsistent results across to be as reliable and precise as possible low- and middle-income countries in the
studies made summarizing and inter- to avoid missing real efects or being International Clinical Trials Registry Plat-
preting the evidence diicult. misled by seemingly positive indings. form. We hope that more and better evalu-
he most rigorous African study hus, trials should be conducted in set- ations will emerge in future updates of our
reported a moderate increase in insti- tings where a suiciently large number Cochrane review to enable us to formulate
tutional deliveries, from around 35% of facilities or geographical areas can be forthright conclusions about the efects of
to 42% (i.e. 7 percentage points; 95% randomized. performance-based inancing in low- and
conidence interval, CI: 1–14).4 Find- Sometimes RCTs are too expensive, middle-income countries. ■
ings from studies in Burundi, 5 the time-consuming or impractical. he
Democratic Republic of the Congo 6 most recent study of performance-based Competing interests: ITO is employed by
and Rwanda7 showed disparate indings: inancing, from Rwanda, illustrates the the Norwegian Agency for Develop-
one reported a signiicant increase in practical setbacks researchers some- ment Cooperation (Norad), which funds
institutional deliveries, another found times face. he study was set up as a performance-based inancing projects in
little or no change and the third showed well-designed RCT with districts ran- several countries. Also, the Norwegian
a signiicant decrease. For these indings domly assigned to performance-based government is one of the contributors to
the risk of bias is high, partly because inancing intervention and non-inter- he World Bank’s Health Results Innova-
intervention and control areas were not vention groups,4 but right before the tion Trust Fund (HRITF).
randomly allocated and the same people trial began, the government redeined
who implemented the programmes the administrative district boundaries References
also evaluated them. Two additional and several of the districts in the control Available at: http://www.who.int/bulletin/
studies reporting on institutional de- group were shited to the intervention volumes/90/8/12-106468
liveries were programme evaluations group. his may have weakened the
with a substantial risk of bias due to validity of the study indings.
a
Norwegian Knowledge Centre for the Health Services, PO Box 7004, St Olavs plass, Oslo, 0130, Norway.
b
Immpact, University of Aberdeen, Aberdeen, Scotland.
c
Norwegian Knowledge Centre for the Health Services, Oslo, Norway.
d
Norwegian Agency for Development Cooperation, Oslo, Norway.
Correspondence to Atle Fretheim (e-mail: ).
Bull World Health Organ 2012;90:559–559A | doi:10.2471/BLT.12.106468 559
