From adjustment to coverage

Coverage is a count of services divided by the population that should have received them. Both sides need work: the counts are adjusted for what was not reported, and the population (the denominator) can be estimated in several ways. This guide follows the steps with the dataset object, the CacheConnection, which remembers each choice and computes the results from them.

library(cd2030.core)

cache <- load_cache_data(system.file("extdata", "kenya.xlsx", package = "cd2030.core"))
#> ℹ Loading Excel "kenya.xlsx" for parts
#> ! The following columns are empty: anc_1trimester, under5_deaths, total_deaths, ipv2, measles_cases
#> ℹ Successfully loaded and cleaned data
#> ℹ Auto-detect candidates (|group|): vaccine=19, rmncah=28 -> selected rmncah

1. Adjust the counts

Adjustment does three things to the monthly counts:

Scales them up for incomplete reporting. With a reporting rate of r percent, a count becomes count * (1 + (100 / r - 1) * k). The factor k, between 0 and 1, says how much service the facilities that did not report are assumed to have given, compared with those that did: 0 means none, 1 means as much.
Replaces extreme outliers (a month more than five median absolute deviations from the district’s median) with the district’s median for the year.
Fills missing values with the district’s median for the year.

There is one k for each group of indicators.

cache$set_k_factors(c(anc = 0.25, idelv = 0.25, vacc = 0.25, opd = 0.25, ipd = 0.25))
cache$adjust_data()

The adjusted data is kept beside the original, and adjustment_values shows what the adjustment changed, year by year.

changes <- cache$adjustment_values
changes[, c("year", "penta1_raw", "penta1_adj", "anc1_raw", "anc1_adj")]
#> # Table 1:  Unadjusted and adjusted values
#> # A tibble: 5 × 5
#>    year penta1_raw penta1_adj anc1_raw anc1_adj
#>   <int>      <dbl>      <dbl>    <dbl>    <dbl>
#> 1  2019    1359967   1380255. 1429512  1451735.
#> 2  2020    1323091   1345884. 1465266  1491785.
#> 3  2021    1411557   1424787. 1547683. 1553043.
#> 4  2022    1387864   1398699. 1479978. 1490510.
#> 5  2023    1410586   1412664. 1500341  1510589.

A year can be left out altogether with cache$set_excluded_years(). Without the dataset object, the same adjustment is adjust_service_data(data, adjustment = "custom", k_factors = ...).

2. Compare the denominators

The populations come from two sources: the projections in the facility data’s population sheet (called dhis2) and the UN estimates (un).

populations <- cache$denominator_metrics
populations[, c("year", "un_births", "totlivebirths_dhis2", "un_under1", "totunder1_dhis2")]
#> # Table 2a: Total Population and Live Birth by Source
#> # Country:  Kenya
#> # A tibble: 5 × 5
#>    year un_births totlivebirths_dhis2 un_under1 totunder1_dhis2
#>   <dbl>     <dbl>               <dbl>     <dbl>           <dbl>
#> 1  2019     1445.               1694.     1407.           1602.
#> 2  2020     1451.               1506.     1407.           1459.
#> 3  2021     1469.               1642.     1419.           1509.
#> 4  2022     1483.               1616.     1436.           1563.
#> 5  2023     1500.               1199.     1453.           1602.

plot(populations, metric = "births")

Two more denominators are derived from the service counts themselves. Where nearly every pregnant woman attends a first antenatal visit, or nearly every infant receives a first dose of pentavalent vaccine, the count divided by the survey coverage of that service estimates the population.

cd_denominator_labels()
#>                         un                      dhis2                       anc1 
#>           "UN projections"        "DHIS2 projections"             "ANC1-derived" 
#>                     penta1                anc1derived              penta1derived 
#>           "Penta1-derived"   "ANC1 population growth" "Penta1 population growth"

3. Give the national rates

The derived denominators step from one population to the next (pregnancies to births to surviving infants) with a few national rates. The package fills in those the country’s latest survey has.

str(cache$national_estimates)
#> List of 7
#>  $ nmr      : num 0.023
#>  $ pnmr     : num 0.0121
#>  $ twin_rate: num 0.015
#>  $ preg_loss: num 0.03
#>  $ sbr      : logi NA
#>  $ anc1     : num 0.98
#>  $ penta1   : num 0.971

Every rate is needed. This survey has no stillbirth rate, so it is set here; the others can be replaced the same way.

cache$set_national_estimates(list(sbr = 0.02))
cache$check_inequality_params
#> [1] TRUE

check_inequality_params and check_coverage_params say whether the dataset has what the coverage calculations need. When one is FALSE, something is missing: adjusted data, a survey year, a national rate, or the UN, WUENIC or survey estimates.

4. Compute coverage

indicator_coverage_national has the populations and coverage of every indicator under every denominator. Columns are named cov_<indicator>_<denominator>.

coverage <- cache$indicator_coverage_national
coverage[, c("year", "cov_penta3_dhis2", "cov_penta3_un", "cov_penta3_anc1", "cov_penta3_penta1")]
#> # Table:    National Coverage Estimates
#> # A tibble: 5 × 5
#>    year cov_penta3_dhis2 cov_penta3_un cov_penta3_anc1 cov_penta3_penta1
#>   <dbl>            <dbl>         <dbl>           <dbl>             <dbl>
#> 1  2019             79.2          92.9            94.6              92.2
#> 2  2020             87.8          91.2            90.7              93.2
#> 3  2021             84.2          94.1            91.1              92.1
#> 4  2022             85.6          93.2            94.9              93.8
#> 5  2023            116.           93.0            94.5              93.7

plot(coverage, plot_type = "anc_coverage_penta1")

The denominators do not agree, which is why they are compared. A coverage above 100, as the facility projections give for 2023, means that denominator is too small for the services counted.

The same tables exist by region and by district: indicator_coverage_admin1, indicator_coverage_district.

5. Choose a denominator

The denominator to use is the one whose coverage agrees best with the surveys. denominator is the one chosen for the dataset, and maternal_denominator the one for maternal indicators.

cache$denominator
#> [1] "penta1"
cache$set_denominator("penta1")

6. Compare with surveys and WUENIC

get_filtered_coverage() takes one indicator, with the chosen denominator, and sets the facility-data estimate beside the survey estimates, their confidence interval, and the WUENIC estimates.

penta3 <- cache$get_filtered_coverage("penta3", "national")
penta3
#> # A tibble: 5 × 7
#>   estimates        `2014` `2019` `2020` `2021` `2022` `2023`
#>   <chr>             <dbl>  <dbl>  <dbl>  <dbl>  <dbl>  <dbl>
#> 1 DHIS2 estimate     NA     92.2   93.2   92.1   93.8   93.7
#> 2 Survey estimates   89.9   NA     NA     NA     89.2   NA  
#> 3 95% CI LL          88.5   NA     NA     NA     87.8   NA  
#> 4 95% CI UL          91.1   NA     NA     NA     90.4   NA  
#> 5 WUENIC estimates   NA     91     89     91     90     93

plot(penta3)

Beyond coverage

The dataset object computes the other analyses the same way. Each is described in ?CacheConnection and under its heading in the reference.

mortality <- cache$mortality_summary
plot(mortality, indicator = "sbr_inst")

Analysis	From the dataset object
Sub-national inequality	`get_filtered_inequality()`, `inequality_admin1`, `inequality_district`
Institutional mortality	`mortality_summary`, `mortality_ratios`
Service utilization	`service_utilization_national`, `service_utilization_admin1`
Health system	`health_system_metrics_national`, `health_system_comparison`
Family planning	`fpet_data`

Write a report

report_presets() are Countdown’s standard reports. export_report() writes one for a dataset, as Word, PDF or a PowerPoint deck.

names(report_presets("en"))
#>  [1] "data_quality"          "adjustment"            "denominator_selection" "national_coverage"    
#>  [5] "national_inequality"   "mortality"             "service_utilization"   "health_system"        
#>  [9] "private_sector"        "synthesis"             "one_pager"

reports <- report_presets("en")
export_report(cache, reports[["national_coverage"]], "national-coverage.docx", format = "docx")

Save the dataset

cache$set_cache_path("kenya.rds")

From then on the dataset saves itself after each change, and load_cache_data("kenya.rds") opens it with the adjustment, rates and denominators set here.