DETERMINANTS OF FORECAST ACCURACY FOR PAEDIATRIC ANTIRETROVIRAL DRUGS IN KENYA NJOGO, SUSAN MUTHONI GICHUKI (U51/82556/2012)

Outline Introduction Study objectives Methodology Results Conclusion Recommendations

Introduction (1) Forecasting is the process of estimating quantity required of particular products to meet demand for a specified period of time. It is a fundamental and basic input to decision making in operations management on future demand, budget preparation, supply planning and storage requirements.

Introduction (2) Two important aspects of forecast are expected level of demand and degree of accuracy Forecast accuracy is a function of the ability of forecasters to correctly model demand, random variation, and sometimes unforeseen events.

Introduction (3) Forecast errors should be stated in order to help forecast users to plan for possible errors and provide basis for comparing alternative forecasts

Introduction (3) Management of HIV infection calls for uninterrupted supply of required products Forecast for both adult and paeds antiretroviral (ARV) drugs is done annually by NASCOP Forecasting is largely dependent on assumptions due to poor quality or lack of required data

Introduction (4) All children <15 years are generally assumed to weigh <25Kg and to use paeds ARV formulations Forecasting for paeds ARV drugs has been reported to be specifically challenging because; – Dosages change over time as a child grows – Dosing is individualized based on weight – A mix of formulations exist for children

Study aim The aim of the study was to evaluate forecast accuracy for paediatric antiretroviral drugs in Kenya and assess determinants of forecast accuracy

Study objectives 1.Establish forecast accuracy for paediatric ARV drugs by for period July 2010 to June Determine the relative proportions of children within various weight categories, ARV regimen and formulations at Mbagathi District Hospital. 3.Determine within and between-individual changes over time with regards to weight, regimen and formulation. 4.Identify factors that influence choice of antiretroviral formulations dispensed to children at Mbagathi District Hospital from the pharmacy staff perspective

Methodology Forecast accuracy was established by computing 12- months and quarterly Mean Absolute Percentage Error (MAPE) using Microsoft Office Excel A retrospective longitudinal cohort design was used to determine effect of age, weight and sex on ARV formulations dispensed to children aged <15 years at Mbagathi District Hospital. In-depth interviews were carried out on pharmacy staff dispensing ARV drugs at Mbagathi District Hospital

RESULTS

Baseline characteristics (1) Characteristic N (%) or median (IQR) July 2010July 2011July 2012 Sex:Male 171(55.0)171(55.9)169(55.8) Female 140(45.0)135(44.1)126(44.2) Age (years) 9.2(6.8,12.0)9.6(7.3, 12.0)10.3(7.5,12.5) Weight (Kg) 26(20,32)27(21, 32)28(21, 34) Weight category: < 25 Kg 140(45.0)118(38.5)100(35.1) ≥25 Kg 171(55.0)188(61.5)185(64.9) ARV formulation: Adult formulation 257(82.6)223(72.9)182(63.9) Paed formulation 54(17.4)83(27.1)103(36.1)

Baseline characteristics (2) Characteristic N (%) or median (IQR) July 2010July 2011July 2012 Regimen d4t/3TC/NVP 189(60.8)153(50.0)70(24.6) AZT/3TC/NVP 77(24.8)87(28.4)131(46.0) ABC/3TC/NVP 21(6.8)36(11.8)46(16.1) ARV formulations d4T/3TC/NVP 30/150/200mg 189(60.8)147(48.0)70(24.6) AZT/3TC/NVP 300/150/200mg 32(10.3)39(12.7)68(23.9) AZT/3TC/NVP 60/30/50mg 25(8.0)47(15.4)62(21.8) ABC/3TC 60/30mg + NVP 200mg 10(3.2)18(5.9)24(8.4)

Forecast accuracy a.Mean Absolute Percentage Error (MAPE) 12-moths MAPE 12-month and quarterly MAPE were observed across all the products However, NVP-10mg/ml had reasonable forecast based on Lewis MAPE scale of judgment of forecast errors

ADF test Run test *z-critical= at 95% confidence level; **significant randomness at p<0.05; Null hypothesis for ADF test: Errors are non-random. * r: >12 and <26;**significant non-randomness at p<0.05; Null hypothesis for Run test: Forecast errors are random. b. Randomness of forecast errors test

Visual inspection for randomness of forecast errors

Univariate analysis a.Proportions of children in the <25Kg and ≥25 Kg weight categories More than 50% of children were in the ≥25 Kg weight category across the three periods Period Kg Kg Kg Kg≥25Kg 2010/110.0%2.80%16.40%21.90%58.90% 2011/120.30%2.70%14.40%17.20%65.40% 2012/130.40%2.30%12.30%16.50%68.50%

b. Average proportion of children for selected ARV combination A significant proportion of children were on adult ARV formulation ARV drug formulation2010/112011/122012/13 d4T/3TC/NVP 30/150/200mg100.0%94.9%100.0% d4T/3TC/NVP 12/60/100mg 0.0%5.1% 0.0% AZT/3TC/NVP 60/30/50mg45.6%47.5%41.1% AZT/3TC/NVP 300/150/200mg44.3%52.3%58.7% ABC/3TC 60/30mg + NVP 200mg59.8%54.4%50.6% ABC 300mg + 3TC 150mg + NVP 200mg37.8%42.5%49.4%

Bivariate analysis a.Relationship between ARV formulation and weight category Chi square test revealed that significant relationship (p=0.00) existed between ARV formulation and weight category across all the years ARV N (%) PeriodFormulation <25 Kg ≥25 Kg P value 2010/11Adult820( 55.0)2113(99.2)0.00* Paediatric671(45.0)18(0.84) 2011/12Adult182 (15.0 )2,261 (98.7)0.00* Paediatric1,033(85.0)30 (1.3) 2012/13Adult30 (2.9 )2,156 (97.0)0.00* Paediatric990 (97.1)66 (3.0)

b. Relationship between age category and weight Profile for mean weight over age

Within-subject effects and population-average response Within-subject effects: Only 17(5.5%), 49(16.0%) and 26(9.1%) of children changed ARV formulation due to weight change for 2010/11, 2011/12 and 2012/13 respectively. Population-average response: Odds ratio of being given paeds ARV formulation when weight is ≥25Kg was 0.37, 0.03 and 0.07 for 2010/11, 2011/12 and 2012/13 respectively Therefore, 63%, 97% and 93% of children were likely to receive adult formulations when weight changed from <25Kg to ≥25Kg for 2010/11, 2011/12 and 2012/13 respectively

In-depth interviews Four factors were identified to influence choice of ARV formulation dispensed to a children; – Weight – age – availability of paediatric ARV drugs and – preference of the dispensing staff to dispense paediatric antiretroviral drugs to children below 15 years regardless of the weight

Conclusion Of the seven paediatric ARV drugs, only nevirapine10mg/ml had reasonably accurate forecasts. However, forecast errors for all drugs were non-randomness. The assumptions that all children aged less than 15 years weigh <25Kg and use paediatric ARV formulations were incorrect. Weight was found to be an important determinant of formulation selection. Recommendation To improve forecast accuracy, Forecast accuracy need to be routinely monitored and evaluated Children weight need to be incorporated in the routinely reported program data item or analysis be carried for a sample of facilities.

Acknowledgement National AIDS/STI Control Program (NASCOP) Mbagathi District Hospital My supervisors – Dr. George Osanjo – Dr. Eric Guantai