Introduction
This vignette demonstrates how to use the mwana package’s functions to estimate the prevalence of wasting. The package allow users to estimate prevalence based on:
- Weight-for-height z-score (WFHZ) and/or edema;
- Raw MUAC values and/or edema;
- MUAC-for-age z-score (MFAZ) and/or edema, and
- Combined prevalence.
The prevalence functions in mwana were carefully conceived and designed to simplify the workflow of a nutrition data analyst, especially when dealing with data sets containing imperfections that require additional layers of analysis. Let’s try to clarify this with two scenarios that I believe will remind you of the complexity involved:
When analysing a multi-area data set, users will likely need to estimate the prevalence for each area individually. Afterward, they must extract the results and collate in a summary table to share.
When working with MUAC data, if age ratio test is rated as problematic, an additional tool is required to weight the prevalence and correct for age bias, thus the associated likely overestimation of the prevalence. In unfortunate cases where multiple areas face this issue, the workflow must be repeated several times, making the process cumbersome and highly error-prone 😬.
With mwana you no longer have to worry about this 🥳 as the functions are designed to deal with that. To demonstrate their use, we will use different data sets containing some imperfections alluded above:
-
anthro.02: a survey data with survey weights. Read more about this data with?anthro.02. -
anthro.03: district-level SMART surveys with two districts whose WFHZ standard deviations are rated as problematic while the rest are within range. Do?anthro.03for more details. -
anthro.04: a community-based sentinel site data. The data has different characteristics that require different analysis approaches.
Now we can begin delving into each function.
Estimation of the prevalence of wasting based on WFHZ
To estimate the prevalence of wasting based on WFHZ we use the mw_estimate_prevalence_wfhz() function. The data set to supply must have been wrangled by mw_wrangle_wfhz().
As usual, we start off by inspecting our data set:
tail(anthro.02)#> # A tibble: 6 × 14
#> province strata cluster sex age weight height edema muac wtfactor wfhz
#> <chr> <chr> <int> <dbl> <dbl> <dbl> <dbl> <chr> <dbl> <dbl> <dbl>
#> 1 Nampula Urban 285 1 59.5 13.8 90.7 n 149 487. 0.689
#> 2 Nampula Rural 234 1 59.5 17.2 105. n 193 1045. 0.178
#> 3 Nampula Rural 263 1 59.6 18.4 100 n 156 952. 2.13
#> 4 Nampula Rural 257 1 59.7 15.9 100. n 149 987. 0.353
#> 5 Nampula Rural 239 1 59.8 12.5 91.5 n 135 663. -0.722
#> 6 Nampula Rural 263 1 60.0 14.3 93.8 n 142 952. 0.463
#> # ℹ 3 more variables: flag_wfhz <dbl>, mfaz <dbl>, flag_mfaz <dbl>We can see that the data set contains the required variables for a WFHZ prevalence analysis, including for a weighted analysis. This data set has already been wrangled, so we do not need to call the WFHZ wrangler in this case. We will begin the demonstration with an unweigthed analysis - typical of SMART surveys - and then we proceed to a weighted analysis.
Estimation of unweighted prevalence
To achieve this we do:
anthro.02 |>
mw_estimate_prevalence_wfhz(
wt = NULL,
edema = edema,
.by = NULL
)This will return:
#> # A tibble: 1 × 16
#> gam_n gam_p gam_p_low gam_p_upp gam_p_deff sam_n sam_p sam_p_low sam_p_upp
#> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 121 0.0408 0.0322 0.0494 Inf 43 0.00664 0.00273 0.0106
#> # ℹ 7 more variables: sam_p_deff <dbl>, mam_n <dbl>, mam_p <dbl>,
#> # mam_p_low <dbl>, mam_p_upp <dbl>, mam_p_deff <dbl>, wt_pop <dbl>If for some reason the variable edema is not available in the data set, or it’s there but not plausible, we can exclude it from the analysis by setting the argument edema to NULL:
anthro.02 |>
mw_estimate_prevalence_wfhz(
wt = NULL,
edema = NULL, # Setting edema to NULL
.by = NULL
)And we get:
#> # A tibble: 1 × 16
#> gam_n gam_p gam_p_low gam_p_upp gam_p_deff sam_n sam_p sam_p_low sam_p_upp
#> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 107 0.0342 0.0263 0.0420 Inf 29 0 0 0
#> # ℹ 7 more variables: sam_p_deff <dbl>, mam_n <dbl>, mam_p <dbl>,
#> # mam_p_low <dbl>, mam_p_upp <dbl>, mam_p_deff <dbl>, wt_pop <dbl>If we inspect the gam_n and gam_p columns of this output table and the previous, we notice differences in the numbers. This occurs because edema cases were excluded in the second implementation. Note that you will observed a change if there are positive cases of edema in the data set; otherwise, setting edema = NULL will have no effect whatsoever.
The above output summary does not show results by province. We can control that using the .by argument. In the above examples, it was set to NULL; now let’s pass the name of the column containing the locations where the data was collected. In our case, the column is province:
anthro.02 |>
mw_estimate_prevalence_wfhz(
wt = NULL,
edema = edema,
.by = province # province is the variable's name holding data on where the survey was conducted.
)And voila :
#> # A tibble: 2 × 17
#> province gam_n gam_p gam_p_low gam_p_upp gam_p_deff sam_n sam_p sam_p_low
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Zambezia 41 0.0290 0.0195 0.0384 Inf 10 0.00351 0.0000639
#> 2 Nampula 80 0.0546 0.0397 0.0695 Inf 33 0.0103 0.00282
#> # ℹ 8 more variables: sam_p_upp <dbl>, sam_p_deff <dbl>, mam_n <dbl>,
#> # mam_p <dbl>, mam_p_low <dbl>, mam_p_upp <dbl>, mam_p_deff <dbl>,
#> # wt_pop <dbl>A table with two rows is returned with each province’s statistics.
Estimation of weighted prevalence
To get the weighted prevalence, we make the use of the wt argument. We pass to it the column name containing the final survey weights. In our case, the column name is wtfactor:
anthro.02 |>
mw_estimate_prevalence_wfhz(
wt = wtfactor, # Passing the wtfactor to wt
edema = edema,
.by = province
)And you get:
#> # A tibble: 2 × 17
#> province gam_n gam_p gam_p_low gam_p_upp gam_p_deff sam_n sam_p sam_p_low
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Zambezia 41 0.0261 0.0161 0.0361 1.16 10 0.00236 -0.000255
#> 2 Nampula 80 0.0595 0.0410 0.0779 1.52 33 0.0129 0.00272
#> # ℹ 8 more variables: sam_p_upp <dbl>, sam_p_deff <dbl>, mam_n <dbl>,
#> # mam_p <dbl>, mam_p_low <dbl>, mam_p_upp <dbl>, mam_p_deff <dbl>,
#> # wt_pop <dbl>The work under the hood of
mw_estimate_prevalence_wfhzUnder the hood, before starting the prevalence estimation, the function first checks the quality of the WFHZ standard deviation. If it is not rated as problematic, it proceeds with a complex sample-based analysis; otherwise, prevalence is estimated applying the PROBIT method. This is as you see in the body of the plausibility report generated by ENA. The
anthro.02data set has no such issues, so you don’t seemw_estimate_prevalence_wfhzin action on this regard. To see that, let’s use theanthro.03data set.
anthro.03 contains problematic standard deviation in Metuge and Maravia districts, while the remaining districts are within range.
Let’s inspect our data set:
#> # A tibble: 6 × 9
#> district cluster team sex age weight height edema muac
#> <chr> <int> <int> <chr> <dbl> <dbl> <dbl> <chr> <int>
#> 1 Metuge 2 2 m 9.99 10.1 69.3 n 172
#> 2 Metuge 2 2 f 43.6 10.9 91.5 n 130
#> 3 Metuge 2 2 f 32.8 11.4 91.4 n 153
#> 4 Metuge 2 2 f 7.62 8.3 69.5 n 133
#> 5 Metuge 2 2 m 28.4 10.7 82.3 n 143
#> 6 Metuge 2 2 f 12.3 6.6 69.4 n 121Now let’s apply the prevalence function. This is data is not wrangled, so we will have to wrangle it before passing to the prevalence function:
anthro.03 |>
mw_wrangle_wfhz(
sex = sex,
.recode_sex = TRUE,
height = height,
weight = weight
) |>
mw_estimate_prevalence_wfhz(
wt = NULL,
edema = edema,
.by = district
)The returned output will be:
#> ================================================================================
#> # A tibble: 4 × 17
#> district gam_n gam_p gam_p_low gam_p_upp gam_p_deff sam_n sam_p sam_p_low
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Metuge NA 0.0251 NA NA NA NA 0.00155 NA
#> 2 Cahora-Ba… 25 0.0738 0.0348 0.113 Inf 4 0.00336 -0.00348
#> 3 Chiuta 11 0.0444 0.0129 0.0759 Inf 2 0.00444 -0.00466
#> 4 Maravia NA 0.0450 NA NA NA NA 0.00351 NA
#> # ℹ 8 more variables: sam_p_upp <dbl>, sam_p_deff <dbl>, mam_n <dbl>,
#> # mam_p <dbl>, mam_p_low <dbl>, mam_p_upp <dbl>, mam_p_deff <dbl>,
#> # wt_pop <dbl>Can you spot the differences? 😎 Yes, you’re absolutely correct! While in Cahora-Bassa and Chiúta districts all columns are populated with numbers, in Metuge and Maravia, only the gam_p, sam_p and mam_p columns are filled with numbers, and everything else with NA. These are district where the PROBIT method was applied, while in Cahora-Bassa and Chiúta ditricts the standard complex sample analysis was done.
Estimation of the prevalence of wasting based on MFAZ
The prevalence of wasting based on MFAZ can be estimated using the mw_estimate_prevalence_mfaz() function. This function works and is implemented the same way as demonstrated in Section 1.1, with the exception of the data wrangling that is based on MUAC. This was demonstrated in the plausibility checks. In this way, to avoid redundancy, we will not demonstrate the workflow.
Estimation of the prevalence of wasting based on raw MUAC values
This job is assigned to mw_estimate_prevalence_muac(). Once you call the function, before starting the prevalence estimation, it first evaluates the acceptability of the MFAZ standard deviation and the age ratio test. Yes, you read well, MFAZ’s standard deviation, not on the raw values MUAC.
Important
Although the acceptability is evaluated on the basis of MFAZ, the actual prevalence is estimated on the basis of the raw MUAC values. MFAZ is also used to detect outliers and flag them to be excluded from the prevalence analysis.
The MFAZ standard deviation and the age ratio test results are used to control the prevalence analysis flow in this way:
- If the MFAZ standard deviation and the age ratio test are both not problematic, a standard complex sample-based prevalence is estimated.
- If the MFAZ standard deviation is not problematic but the age ratio test is problematic, the SMART MUAC tool age-weighting approach is applied.
- If the MFAZ standard deviation is problematic, even if age ratio is not problematic, no prevalence analysis is estimated, instead
NAare thrown.
When working with a multiple-area data set, these conditionals will still be applied according to each area’s situation.
How does it work on a multi-area data set
Fundamentally, the function performs the standard deviation and age ratio tests, evaluates their acceptability, and returns a summarized table by area. It then iterates over that summary table row by row checking the above conditionals. Based on the conditionals of each row (area), the function accesses the original data set, computes the prevalence accordingly, and returns the results.
To demonstrate this we will use the anthro.04 data set.
As usual, let’s first inspect it:
#> # A tibble: 6 × 8
#> province cluster sex age muac edema mfaz flag_mfaz
#> <chr> <int> <dbl> <int> <dbl> <chr> <dbl> <dbl>
#> 1 Province 3 743 2 21 130 n -1.50 0
#> 2 Province 3 743 2 9 126 n -1.33 0
#> 3 Province 3 743 2 12 128 n -1.27 0
#> 4 Province 3 743 2 34 145 n -0.839 0
#> 5 Province 3 743 2 11 130 n -1.04 0
#> 6 Province 3 743 2 33 140 n -1.23 0You see that this data has already been wrangled, so we will go straight to the prevalence estimation.
Important
As in ENA Software, make sure you run the plausibility check before you call the prevalence function. This is good to know about the acceptability of your data. If we do that with
anthro.04we will see which province has issues, hence what we should be expecting to see in below demonstrations based on the conditionals stated above.
anthro.04 |>
mw_estimate_prevalence_muac(
wt = NULL,
edema = edema,
.by = province
)This will return:
#> # A tibble: 3 × 17
#> province gam_n gam_p gam_p_low gam_p_upp gam_p_deff sam_n sam_p sam_p_low
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Province 1 135 0.104 0.0778 0.130 Inf 19 0.0133 0.00682
#> 2 Province 2 NA 0.112 NA NA NA NA 0.0201 NA
#> 3 Province 3 NA NA NA NA NA NA NA NA
#> # ℹ 8 more variables: sam_p_upp <dbl>, sam_p_deff <dbl>, mam_n <dbl>,
#> # mam_p <dbl>, mam_p_low <dbl>, mam_p_upp <dbl>, mam_p_deff <dbl>,
#> # wt_pop <dbl>We see that in Province 1, all columns are filled with numbers; in Province 2, some columns are filled with numbers, while other columns are filled with NAs: this is where the age-weighting approach was applied. Lastly, in Province 3 a bunch of NA are filled everywhere - you know why 😉 .
Alternatively, we can choose to apply the function that calculates the age-weighted prevalence estimates inside mw_estimate_prevalence_muac() directly on to our data set. This can be done by calling the mw_estimate_smart_age_wt() function. It worth noting that although possible, we recommend to use the main function. This is simply due the fact that if we decide to use the function independently, then we must, before calling it, check the acceptability of the standard deviation of MFAZ and of the age ratio test, and then evaluate if the conditions that fits the use mw_estimate_smart_age_wt() are there. We would have to do that ourselves. This introduces some kind of cumbersomeness in the workflow, and along with that, a risk of picking a wrong analysis workflow.
Nonetheless, if for any reason we decide to go for it anyway, then we would apply the function as demonstrated below. We will continue using the anthro.04 data set. For this demonstration, we will just pull out the data set from Province 2 where we already know that the conditions to apply mw_estimate_smart_age_wt() are met, and then we will pipe it in to the function:
anthro.04 |>
subset(province == "Province 2") |>
mw_estimate_smart_age_wt(
edema = edema,
.by = NULL
)This returns the following:
#> # A tibble: 1 × 3
#> sam_p mam_p gam_p
#> <dbl> <dbl> <dbl>
#> 1 0.0201 0.0922 0.112Estimation of weighted prevalence
For this we go back anthro.02 data set.
We approach this task as follows:
## Load library ----
library(dplyr)
## Compute prevalence ----
anthro.02 |>
mw_wrangle_age(
age = age,
.decimals = 2
) |>
mw_wrangle_muac(
sex = sex,
.recode_sex = FALSE,
muac = muac,
.recode_muac = TRUE,
.to = "cm",
age = "age"
) |>
mutate(
muac = recode_muac(muac, .to = "mm")
) |>
mw_estimate_prevalence_muac(
wt = wtfactor,
edema = edema,
.by = province
)This will return:
#> ================================================================================
#> # A tibble: 2 × 17
#> province gam_n gam_p gam_p_low gam_p_upp gam_p_deff sam_n sam_p sam_p_low
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Nampula 70 0.0571 0.0369 0.0773 2.00 28 0.0196 0.00706
#> 2 Zambezia 65 0.0552 0.0380 0.0725 1.67 18 0.0133 0.00412
#> # ℹ 8 more variables: sam_p_upp <dbl>, sam_p_deff <dbl>, mam_n <dbl>,
#> # mam_p <dbl>, mam_p_low <dbl>, mam_p_upp <dbl>, mam_p_deff <dbl>,
#> # wt_pop <dbl>Warning
You may have noticed that in the above code block, we called the
recode_muac()function insidemutate(). This is because after you usemw_wrangle_muac(), it puts the MUAC variable in centimeters. Themw_estimate_prevalence_muac()function was defined to accept MUAC in millimeters. Therefore, it must be converted to millimeters.
Estimation for non survey data
Thus far, the demonstration has been around survey data. However, it is also common in day-to-day practice to come across non survey data to analyse. Non survey data can be screenings or any other kind of community-based surveillance data. With this kind of data, the analysis workflow usually consists in a simple estimation of the point prevalence and the counts of the positive cases, without necessarily estimating the uncertainty. mwana provides a handy function for this task: mw_estimate_prevalence_screening(). Under the hood, this function works exactly the same way as mw_estimate_prevalence_muac(). The only difference is that it was designed to deal with non survey data.
To demonstrate its usage, we will use the anthro.04 data set.
anthro.04 |>
mw_estimate_prevalence_screening(
muac = muac,
edema = edema,
.by = province
)The returned output is:
#> # A tibble: 3 × 7
#> province gam_n gam_p sam_n sam_p mam_n mam_p
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Province 1 133 0.104 17 0.0133 116 0.0908
#> 2 Province 2 NA 0.112 NA 0.0201 NA 0.0922
#> 3 Province 3 NA NA NA NA NA NAEstimation of the combined prevalence of wasting
The estimation of the combined prevalence of wasting is a task attributed to the mw_estimate_prevalence_combined() function. The case-definition is based on the WFHZ, the raw MUAC values and edema. From the workflow standpoint, it combines the workflow demonstrated in Section 1.1 and in Section 1.3.
To demonstrate it’s implementation we will use the anthro.01 data set.
Let’s inspect the data:
#> # A tibble: 6 × 11
#> area dos cluster team sex dob age weight height edema
#> <chr> <date> <int> <int> <chr> <date> <int> <dbl> <dbl> <chr>
#> 1 District E 2023-12-04 1 3 m NA 59 15.6 109. n
#> 2 District E 2023-12-04 1 3 m NA 8 7.5 68.6 n
#> 3 District E 2023-12-04 1 3 m NA 19 9.7 79.5 n
#> 4 District E 2023-12-04 1 3 f NA 49 14.3 100. n
#> 5 District E 2023-12-04 1 3 f NA 32 12.4 92.1 n
#> 6 District E 2023-12-04 1 3 f NA 17 9.3 77.8 n
#> # ℹ 1 more variable: muac <int>Data wrangling
Fundamentally, it combines the data wrangling workflow of WFHZ and MUAC:
## Load library ----
library(dplyr)
## Apply the wrangling workflow ----
anthro.01 |>
mw_wrangle_age(
dos = dos,
dob = dob,
age = age,
.decimals = 2
) |>
mw_wrangle_muac(
sex = sex,
.recode_sex = TRUE,
muac = muac,
.recode_muac = TRUE,
.to = "cm",
age = "age"
) |>
mutate(
muac = recode_muac(muac, .to = "mm")
) |>
mw_wrangle_wfhz(
sex = sex,
weight = weight,
height = height,
.recode_sex = FALSE
)This is to get the wfhz and flag_wfhz the mfaz and flag_mfaz added to the data set. In the output below, we have just selected these columns:
#> ================================================================================
#> ================================================================================
#> # A tibble: 1,191 × 5
#> area wfhz flag_wfhz mfaz flag_mfaz
#> <chr> <dbl> <dbl> <dbl> <dbl>
#> 1 District E -1.83 0 -1.45 0
#> 2 District E -0.956 0 -1.67 0
#> 3 District E -0.796 0 -0.617 0
#> 4 District E -0.74 0 -1.02 0
#> 5 District E -0.679 0 -0.93 0
#> 6 District E -0.432 0 -1.10 0
#> 7 District E -0.078 0 -0.255 0
#> 8 District E -0.212 0 -0.677 0
#> 9 District E -1.07 0 -2.18 0
#> 10 District E -0.543 0 -0.403 0
#> # ℹ 1,181 more rowsUnder the hood, mw_estimate_prevalence_combined() applies the same analysis approach as in mw_estimate_prevalence_wfhz and in mw_estimate_prevalence_muac(). It checks the acceptability of the standard deviation of WFHZ and MFAZ and of the age ratio test. The following conditionals are checked and applied:
- If the standard deviation of WFHZ and of MFAZ, and the age ratio test are all concurrently not problematic, the standard complex sample-based estimation is applied.
- If any of the above is rated problematic, the prevalence is not computed and
NAs are thrown.
In this function, a concept of “combined flags” is used.
What is combined flag?
Combined flags consists of defining as flag any observation that is flagged in either
flag_wfhzorflag_mfazvectors. A new columncflagsfor combined flags is created and added to the data set. This ensures that all flagged observations from both WFHZ and MFAZ data are excluded from the prevalence analysis.
| flag_wfhz | flag_mfaz | cflags |
|---|---|---|
| 1 | 0 | 1 |
| 0 | 1 | 1 |
| 0 | 0 | 0 |
Now that we understand what happens under the hood, we can now proceed to implement it:
## Load library ----
library(dplyr)
## Apply the workflow ----
anthro.01 |>
mw_wrangle_age(
dos = dos,
dob = dob,
age = age,
.decimals = 2
) |>
mw_wrangle_muac(
sex = sex,
.recode_sex = TRUE,
muac = muac,
.recode_muac = TRUE,
unit = "cm",
.to = "age"
) |>
mutate(
muac = recode_muac(muac, .to = "mm")
) |>
mw_wrangle_wfhz(
sex = sex,
weight = weight,
height = height,
.recode_sex = FALSE
) |>
mw_estimate_prevalence_combined(
wt = NULL,
edema = edema,
.by = area
)We get this:
#> ================================================================================
#> ================================================================================
#> # A tibble: 2 × 17
#> area cgam_n cgam_p cgam_p_low cgam_p_upp cgam_p_deff csam_n csam_p
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 District E NA NA NA NA NA NA NA
#> 2 District G 55 0.0703 0.0447 0.0958 Inf 13 0.00747
#> # ℹ 9 more variables: csam_p_low <dbl>, csam_p_upp <dbl>, csam_p_deff <dbl>,
#> # cmam_n <dbl>, cmam_p <dbl>, cmam_p_low <dbl>, cmam_p_upp <dbl>,
#> # cmam_p_deff <dbl>, wt_pop <dbl>In district E NAs were returned because there were issues with the data. I leave it to you to figure out what was/were the issue/issues.
Tip
Consider running the plausibility checkers.