Naod Mekonnen : Health systems research

1. Economic analysis of ART task shifting in
limited resource setting using econometric
model: Ethiopia Case Study
Elias Asfaw, Naod Mekonnen, Benjamin Johns, Wendy Wong, Abebe
Bekele, Yibeltale Assefa, John Palen, Amha Kebede, Shara Domin
13th Annual International Conference on
Health Economics, Management and Policy 23-26 June 2014,
Athens, Greece

2. Table of Content
 Background ~ ART task shifting
 Antiretroviral treatment program
 Objective of the study
 Methods
 Econometric model
 Results
 Descriptive Analysis
 Econometric Model Analysis
 Conclusion and future research direction

3. Background-ART Task shifting
 Sub-Saharan Africa is facing a crisis in human
health resources.
 Need to be scaled up by as much as 140% to
attain international health development targets
(Kinfu et al, 2009).
 The burden of HIV/AIDS has led to task shifting
 Task shifting- the name now given to a process of
delegation whereby tasks are moved, where
appropriate, to less specialized health workers.
(WHO, 2007)

4. ART Task shifting ~ Ethiopia
 In Ethiopia, the ART delivery model has shifted
from a physician-led hospital based program to
decentralized service delivery at health centers
by the nurses and health officers.
0
200
400
600
800
2004/2005 2010/2011
3
743
ART providing health facility
0
50000
100000
150000
200000
250000
2004/2005 2010/2011
8276
247805
Antiretroviral service users
Source: FHAPCO ART performance report

5. ART Task shifting ~ Ethiopia
 However, there is scanty information available
concerning the cost differences associated with
task shifting and decentralization and
 The determinants of any possible cost
differences.
 The length of visit and type of the health
professional providing the services affects the
magnitude of cost difference.

6. Study Objective
 The research focuses on the relationship
between:
 The length of a visit and the labor cost of visit with
The explanatory variables:
1. Type of health profession type (doctors, health
officers and nurses),
2. Type of health facilities (hospitals, health
centers) and
3. Specific purpose of the visit (initiation, follow-
up).

7. Method
 A stratified random sample of health facilities
across four regions of Ethiopia (Addis Ababa,
Amhara, Benishangul Gumuz, and Oromia) is
included in this study.
 The unit of randomization was the health
network, consisting of one hospital and its
associated health centers.
 The cost of resources used to deliver ART was
collected from the health facilities and the
Federal Ministry of Health.

8. Method…
 Data were collected on the number of staff working in
the ART clinic, how much time they spend in the ART
clinic, how much time they work overall, and staff
salaries.
 Using a time motion study and task checklist, staff roles
and time spent in ART care was assessed.
 The labor cost spent for the visit was calculated and
estimated through multiplying the length of the visit
with average salary of a particular cadre of worker per
minute.
 All costs are presented in 2011 Ethiopian Birr.

9. Econometric Model…
Econometric Model:
Though there were many determinant variables
contributing to ART program success, this study
emphasizes on the above mentioned 3 determinant
variables.
An ordinary least square (OLS) model was applied
with the following equations:
𝑌𝐿𝑖= 𝛼1 + 𝛼2 𝐷2𝑖 + 𝛼3 𝐷3𝑖 + 𝛼4 𝐷4𝑖 + 𝛼5𝑖 𝐷5𝑖 + 𝑈𝑖 … … … … … … … … … … . . 4.1
𝑌𝐶𝑖 = 𝛼1 + 𝛼2 𝐷2𝑖 + 𝛼3 𝐷3𝑖 + 𝛼4 𝐷4𝑖 + 𝛼5𝑖 𝐷5𝑖 + 𝑈𝑖 … … … … … … … . … … … 4.2

10. Econometric Model…
Where:
• YLi= the length of visit,, YCi = Cost of visit,
• D2i= 1 if the professional is Nurse; 0 otherwise:
• D3i= 1 if the professional Health officer; 0 otherwise:
• D4i=1 if the facility is Hospital; 0 if facility is Health
Centre, and
• D5i =1 if the type of patient seen at that particular
visit is initiation; 0 if the type of patient seen at that
particular visit is follow-up.

11. Econometric Model…
Log transformation of the dependent variable and using
generalized linear models.
o For symmetric distribution of the variables, low skewness
value and low kurtosis value
Cost of visit Log of cost of visit Length of visit Log of visit
Number of
observation
665 665 665 665
Mean 1.72 0.35 8.42 1.95
Median (p50) 1.42 0.35 7.00 1.94
SD 1.14 0.61 5.67 0.61
Skewness 2.34 -0.04 2.35 -0.41
Kurtosis 15.1 2.81 14.86 2.80

12. Econometric Model…
Considering the above factors, ordinary least square
(OLS) regression was applied to the log transformed
dependent variable
 To estimate the difference among health professional types,
type of health facility and type of visit.
 Both models were tested for proper functional
form, outliers, residual normality,
homoskedasticity, and multicollinearity.
𝐿𝑛YLi/𝐿𝑛𝑌𝐶𝑖 = 𝛼1 + 𝛼2 𝐷2𝑖 + 𝛼3 𝐷3𝑖 + 𝛼4 𝐷4𝑖 + 𝛼5𝑖 𝐷5𝑖…..4.3

13. Ethical Clearance…
The study approval in:
– the scientific and ethical review office (SERO) of the
Ethiopian Public Health Institute and
– Institutional Review Board of Abt Associates, USA

14. Finding
A total of 79 health facilities covered during the survey.
Time motion studies and task checklists were conducted
with 665 patients.
Descriptive Analysis
o The majority of patients (77%) saw nurses, while 19.6%
and 3.5% were treated by health officers and doctors.
o 62% of the patients visited a health center, while the
remainder (38 percent) attended a hospital
o More than 60% of the patients were examined were
categorized as WHO stage 1 or II

15. Finding…
Length of Visit:
• The average minutes spent by the a patient for ART
services was 8.46 minutes (range 1 to 60 minutes)per visit.
• The average time spent with patients was similar at
health centers (8.3 minutes) and hospital (8.9 minutes).
• The length of a visit with a doctor and senior a nurse
was almost similar, with each average spent 8.7
minutes.
– Health officers spent average 7.6 minutes with a
patient

16. Finding…
Cost of Visit:
o The overall mean cost per visit was 2 ETB, with the
range of 0 to 12 ETB (less than 0.5 USD).
o The variation can be attributed to two factors
1. the amount of salary paid and
2. the amount of time spent during the visit.
The highest visit cost was recorded when a patient visited
a physician/doctor (3.02 ETB), while visits with a nurse
had the lowest cost (1.90 ETB).

17. Finding…
Cost of Visit:
o The labor cost per visit further varies with the types
of the services provided: initiation, follow-up and
referral.
– The highest cost was for patient on initiation (2.32 ETB),
– the lowest was for patients with continuing/ follow-up
visits (1.95 ETB).
o The cost of a visit at hospitals averaged 2.04 ETB,
which is slightly higher than the average visit cost at
health centers (1.91 birr).

18. Finding…
1) Result of model for the log of cost per visit
Variables Coef. Std. Err. P>t
Professional 0.15 0.08 0.02*
Facility 0.01 0.05 0.83
Type of visit -0.01 0.20 0.97
Constant 0.35 0.41 0.39R2 for the above model was found to be 0.02, showing the model does explains
little of the variation in the cost per visit.
The insignificant result from the Cook-Weisburg test
indicates no heteroskadasticy.

19. Finding…
2) Result of model for the log of cost per visit
• The second model was estimated using the log of length of visit
as the dependent variable. The coefficients of the second model
are very similar to first model on the labor cost per visit.
Variables Coef. Std. Err. P>t
Professional
0.15 0.08 0.07
Type of visit
-0.01 0.20 0.97
Facility
0.01 0.05 0.83
Constant 1.94 0.41 0.00

20. Finding…
The interpretation of the above two models follow the
approaches suggested by Halvorson and Palmquist
(2008).
The median length of visit for doctors (D=1) is found to be higher
than that of nurses and health officers by 16 percent when
controlling for health facility type and type of visit (P<0.05)
The median cost per visit in hospitals (D=1) to be 1 percent
higher than that of health centers (p-value = 0.97).
The length of visit for a follow up/ continuing visit (D=1) would
be 1 percent lower than that of an initiation visit (p-value =
0.83).

21. Finding…
Similar results were obtained using the length of
visit as the dependent variable
The model suggests again that the median length of a
visit for doctors (D=1) is longer than that of nurses and
health officers by about 16 percent (p value < 0.05)
The results also indicates minimal differences in the
length of visit in hospitals (D=1) than in health centers
by type of visit

22. Policy Implication from the finding
The study indicated a 16 percent lower length of visit
and labor cost per visit for ART patients treated by
nurses and health officers as compared to physicians.
o These findings support the principles underlying task
shifting (Huicho et al, 2008)
o The study strongly supports the recent momentum of
task shifting initiation by WHO (2007)
Unlike the variable on the health professionals types,
there was no statistically difference detected on the
other variables, which suggests that efficiency gains are
due, in fact, to task shifting.

23. Policy Implication from the finding…
Similar studies also indicated task-shifting as a
cost-effective measure to maximize health
workforce output (Babigumira, Castelnuovo et al.
2009), but should not be offset by cost attributed
to compromised quality of care (Zachariah et al,
2009)
The small estimated visit cost could attribute to
the lower amount of salary payment at the
public health facilities in Ethiopia and huge flow
of ART patients due to economic of scale.

24. Future research recommendation and
limitation of the study…
Future research Limitation of the study
o Further research on other
determinant variables
such as
• quality of care should
be analyzed in order to
show their cost
difference.
o The impact of ART task
shifting remains an
important research
question.
Due to limited availability of
data only 3 determinant
variables considered in our
model
The models were limited to
supply-side independent
variables, without recognizing
other variables (socio
demographic, quality of care)
The cost analysis emphasized
only the labor visit cost

25. Acknowledgment