[External] Taimaka CMAM CEA - Q1 2024 v2

	A	B	D	E	F	L
1	This is an external copy of Taimaka's CEA, diverging from the original as of April 1st, 2024.
2
3	This document is based on GW's modeling of ALIMA's combined protocol CMAM program in Niger and Nigeria. The original model is available here, and our internal working copy is here.
4	Where possible, I have taken inputs from GW's model and applied them here. I chose to do so by copying/pasting numbers, rather than formulas, in order to avoid unpredictable cascading changes we do not understand and get a better feel for the model ourselves.
5	GW only models 6-59 months patients. To match them, I have essentially ignored <6 months patients in our program (excluding them from age %s, etc.). This is on the logic that we could stop treating them if we later discovered they were dragging down CE (though my expectation is that they would probably improve CE).
6
7	General Inputs		Used on another sheet?	Notes	Unfortunately inputs section is currently non-exhaustive. There are model inputs that are not listed here (e.g., program data on other sheets).
8	Key
9	Came from GW CEA					Overall Cost Effectiveness
10	Came from Taimaka program data					20.2
11	Came from Taimaka research
12	Calculation
13						In multiples of GiveDirectly
14	Age at admission breakdown			I'm excluding U6 months cases b/c GW does. For simplicity, just excluding them from modeling for now.
15	6-11 months	36.49%		Mortality + treatment sheet
16	12-23 months	47.66%		Mortality + treatment sheet
17	24-35 months	12.83%		Mortality + treatment sheet
18	36-47 months	2.04%		Mortality + treatment sheet
19	48-59 months	0.98%		Mortality + treatment sheet
20
21	Moral Weights
22	Value of saving a life - post-neonatal (1 - 11 months)	101
23	Value of saving a life - 12-59 months	127		GW calls this category 1 to 4 years, but then lumps in kids up to 59 months in it. I guess they count first year of life as "0 years" and so 4 is really 5?
24	Avg. value of saving a life for a child receiving treatment	117.5		GW calculation.
25
26	Units of value treated from development effects per treatment-year from SMC	0.08
27	Development effects of malnutrition treatment as a percentage of development effects for SMC	171%
28	Units of value from development effects of malnutrition treatment	0.14		GW calculation.
29
30	Costs
31	Cost per child treated, excluding overhead costs	$66.37		Based on our 2024 treatment budget for 4,500 kids, 70-30 SAM/MAM split, exchange rate of N1500.
32	% of overhead costs to include in CEA to simulate scale	45%		Overhead costs as a percentage of total spending should decrease as caseload increases. Thus it is unrealistic to include all of our overhead costs when modeling the program at low scale, as we intend for it to grow, and we typically want this model to provide a sense of our cost-effectiveness at a reasonable scale. For context, this model estimates a 9k patient caseload per year on average. Theorize that modeled overhead costs (which are very rough) could handle a 20k patient program.
33
34	Subjective Inputs
35	Which ceiling analysis values should be used?	Taimaka		Incorporating our data into the ceiling analysis reduces the downward adjustment from the analysis relative to the value used by GW in their analysis of ALIMA. But our coverage/prevalence data, which factors into the ceiling analysis, is somewhat crude.
36	Should all of our overhead be included from the get go, or should it be discounted for our scale?	Discounted		See costs section above - essentially, do you want to know our cost-effectiveness right now or what it will be in a few years with reasonable growth?
37	Should treatment effects be estimated according to Taimaka program data, or GiveWell's estimates?	Taimaka		Our WHZ improvements are substantially higher than what GW models for, but our method of turning those improvements into risk ratio estimates is also somewhat crude.
38	Should government treatment quality be estimated based on Taimaka's knowledge of Gombe, or using GiveWell estimates?	Taimaka		Based on Taimaka's knowledge, GiveWell overestimates quality of government programs, at least when it comes to Gombe.
39
40	Mortality Reductions
41	Without treatment
42	Annual all-cause mortality rate, children 6-59 months, untreated MAM, initial estimate	10.22%		See calculations in 'Mortality + Treatment (Scratch)' sheet. Currently have some reservations about the quality of our coverage + prevalence data.
43	Annual all-cause mortality rate, children 6-59 months, untreated SAM, initial estimate	26.76%		See calculations in 'Mortality + Treatment (Scratch)' sheet. Currently have some reservations about the quality of our coverage + prevalence data.
44	Plausibility adjustment from ceiling analysis	43.20%		See calculations from 'Mortality Ceiling Analysis (Scratch)' sheet. I then add in the same 90% discount for Nigeria that GW does in their sheet. I am unsure of this is warranted in our case, need to finish effect size check.
45
46	Annual all-cause mortality rate, children 6-59 months, untreated MAM,adjusted	4.41%
47	Annual all-cause mortality rate, children 6-59 months, untreated SAM, adjusted	11.56%
48
49	Mortality risk reduction from government malnutrition treatment
50	Relative risk of mortality with MAM treatment	0.60		From Roodman's constructed risk ratios, see table 10 of his paper.
51	Relative risk of mortality with SAM treatment	0.36		From Roodman's constructed risk ratios, see table 10 of his paper.
52	Percentage point reudction in annual mortality among all treated children with MAM, initial estimate	1.77%		Calculation from GW.
53	Percentage point reudction in annual mortality among all treated children with SAM, initial estimate	7.40%		Calculation from GW.
54	Adjustment for Gombe state government treatment quality, SAM	0.40		GW simulates lower treatment quality of government programs by estimating an average WHZ gain of 1.5, based on a set of studies of treatment programs and then a downward adjustment for expected lower government recovery rates. Both of these steps can be found in this spreadsheet. Their estimates of government program quality are too optimistic, at least when it comes to Gombe. I apply an additional adjustment to correct.
55	Adjustment for Gombe state government treatment quality, MAM	0.47
56	Percentage point reudction in annual mortality among all treated children with MAM, adjusted	0.83%
57	Percentage point reudction in annual mortality among all treated children with SAM, adjusted	2.96%
58
59	Mortality risk reduction from Taimaka malnutrition treatment
60	Relative risk of mortality with MAM treatment	0.61		See 'Mortality + Treatment (Scratch)' sheet.
61	Relative risk of mortality with SAM treatment	0.17		See 'Mortality + Treatment (Scratch)' sheet.
62	Percentage point reduction in annual mortality among all treated children with MAM	1.74%		I don't make an adjustment here for treatment quality b/c the avg. recovery rate found in GW's estimation spreadsheet is roughly comparable to us. Actually we're way better for MAM, but for simplicity I don't make an adjustment (seems like it would end up being fairly minor in the scheme of things, though our MAM recovery rate is like 10 pp higher).
63	Percentage point reduction in annual mortality among all treated children with SAM	9.64%
64
65	Number of cases receiving treatment b/c of Taimaka's program
66	MAM cases who would have received no treatment	2578		See 'Mortality and Treatment (scratch)' sheet.
67	SAM cases who would have received no treatment	5689		See 'Mortality and Treatment (scratch)' sheet.
68	MAM cases who would have received gov treatment	122		See 'Mortality and Treatment (scratch)' sheet.
69	SAM cases who would have received gov treatment	611		See 'Mortality and Treatment (scratch)' sheet.
70
71	Number of children receiving treatment b/c of Taimaka's program
72	MAM children who would have received no treatment	2449		See 'Mortality and Treatment (scratch)' sheet.
73	SAM children who would have received no treatment	5404		See 'Mortality and Treatment (scratch)' sheet.
74	MAM children who would have received gov treatment	116		See 'Mortality and Treatment (scratch)' sheet.
75	SAM children who would have received gov treatment	581		See 'Mortality and Treatment (scratch)' sheet.
76
77	Deaths averted
78	Annual number of MAM deaths averted	44		Calculation from GW.
79	Annual number of SAM deaths averted	559		Calculation from GW.
80	IV/EV adjustment - socioeconomic confounding	80%
81	IV/EV adjustment - treatment in historical cohorts	110%
82	IV/EV adjustment - rough categories	110%
83	IV/EV adjustment - lower SAM recovery rates	93%
84	Total - Internal and external validity adjustments	90.02%		Calculation from GW.
85	Annual number of MAM deaths averted, adjusted	39
86	Annual number of SAM deaths averted, adjusted	504
87
88	Units of value
89	Annual units of value from mortality reduction	63,803.19		For reference, quick + dirty model has 37,045 units of value (when adjusted for change in # of patients from modeling caseload over 3 years rather than 1 year).
90
91	Development Effects			Notably, effects of deworming are not included at all, which seems wrong.
92	Adjustment for children changing from government treatment to Taimaka treatment	69%		GW calculation. Much less of a downward adjustment than in the GW CEA b/c gov treatment in our estimation is worse.
93	Units of value per child going from no treatment to Taimaka treatment, developmental effects	0.14		See moral weights inputs.
94	Units of value per child going from gov treatment to Taimaka treatment, developmental effects	0.09		Calculation from GW.
95	Total units of value from development effects	1140.35		Calculation from GW.
96	Internal and external validity adjustment	38.89%		Calculation from GW. Less of an adjustment b/c our ceiling analysis is different b/c includes our data.
97	Adjusted total units of value from development effects	443.49
98
99	Vaccination Benefits			GW models two kinds of vaccination benefits - (1) kids getting vaccinated who otherwise would not have, and (2) kids getting vaccinated earlier than they otherwise would have (but who would have been vaccinated anyway). Our data on our impact on vaccinations unfortunately isn't great, so I don't feel comfortable modeling both of these. Instead, I model just pathway (1), b/c I think it is very reasonable to extrapolate some benefit from our vaccination work in this area from anecdotal evidence + evidence from other NGOs (b/c we do the same things they do).
100	Benefits in children who would not have been vaccinated in the absence of Taimaka's program