Analysis walkthrough: hypertension prevalence

library(chmsflow)
library(recodeflow)
library(dplyr)

Overview

This tutorial walks through a complete analysis using chmsflow: estimating hypertension prevalence for CHMS cycle 3. The same workflow applies to cycles 4–6 – just swap the cycle number.

The steps are:

1. Load cycle data and medication data
2. Recode medications and merge into cycle data
3. Derive hypertension status and its input variables
4. Inspect results
5. Combine cycles into one dataset

1. Load data

chmsflow includes bundled dummy data for all six CHMS cycles. Each cycle has a main dataset (cycle3, cycle4, etc.) and a medication dataset (cycle3_meds, cycle4_meds, etc.).

# Bundled dummy data is loaded automatically with library(chmsflow)
dim(cycle3)

[1] 50 81

dim(cycle3_meds)

[1] 4000    3

At an RDC, you would load real CHMS data from Stata or SAS files instead. See Using chmsflow at an RDC for setup instructions.

2. Recode medications

Medication recoding must happen before deriving hypertension status, because the hypertension derivation uses any_htn_med as an input. For full details on medication recoding, see Recoding medications.

recode_meds_cycles3to6() classifies each medication row by drug class, aggregates to one row per respondent, and merges the result into the main cycle data.

cycle3 <- recode_meds_cycles3to6(cycle3, cycle3_meds, c("any_htn_med", "diab_med"))

After this step, any_htn_med and diab_med are columns in the cycle dataset.

select(cycle3, clinicid, any_htn_med, diab_med) |> head()

  clinicid any_htn_med diab_med
1        1           1        1
2        2           1        1
3        3           1        1
4        4           1        1
5        5           1        1
6        6           1        1

3. Derive hypertension status

Use recode_after_meds() instead of rec_with_table() for variables that depend on medication status. It automatically excludes medication-specific rows from variable_details so the pre-computed medication columns are passed through rather than re-derived.

Hypertension derivation requires a chain of intermediate variables: blood pressure measurements, comorbidities (diabetes, CVD, CKD), and their inputs. List all of them so recode_after_meds() can derive them in order.

htn_variables <- c(
  # Blood pressure (raw + adjusted)
  "bpmdpbps", "bpmdpbpd", "sbp_adj_mmhg", "dbp_adj_mmhg",
  # Medication inputs (merged in step 2)
  "any_htn_med", "ccc_32",
  # Diabetes chain (input to hypertension functions)
  "lab_hba1", "diab_a1c", "ccc_51", "diab_med", "diab_status",
  # CVD chain
  "ccc_61", "ccc_63", "ccc_81", "cvd_status",
  # CKD chain
  "lab_bcre", "pgdcgt", "clc_sex", "clc_age", "gfr_ml_min", "ckd_status",
  # Hypertension outcomes
  "htn_status", "htn_adj_status", "htn_control_status", "htn_control_adj_status"
)

cycle3_htn <- recode_after_meds(cycle3, htn_variables)

NOTE for pgdcgt: Respondents who respond as indigenous to previous question are identified as 'not applicable' in this question. Recode to other"

4. Inspect results

The result contains the derived hypertension variables alongside clinicid. Values are:

• 1 – high blood pressure
• 2 – normal blood pressure
• tagged_na("a") – not applicable
• tagged_na("b") – missing

cycle3_htn |>
  select(clinicid, htn_status, htn_adj_status) |>
  head(10)

   clinicid htn_status htn_adj_status
1         1          1              1
2         2          1              1
3         3          1              1
4         4          1              1
5         5          1              1
6         6          1              1
7         7          1              1
8         8          1              1
9         9          1              1
10       10          1              1

To see the distribution and prevalence:

table(cycle3_htn$htn_status, useNA = "always")

   1 <NA>
  50    0 

cycle3_htn |>
  filter(!is.na(htn_status)) |>
  summarise(
    n = n(),
    n_htn = sum(htn_status == 1),
    prevalence = mean(htn_status == 1)
  )

   n n_htn prevalence
1 50    50          1

For more on how chmsflow handles missing data codes, see Missing data (tagged_na).

5. Combine cycles

To analyse multiple cycles, repeat steps 2–3 for each cycle, add a cycle identifier, and bind them together.

# Repeat for cycles 4-6
cycle4 <- recode_meds_cycles3to6(cycle4, cycle4_meds, c("any_htn_med", "diab_med"))
cycle5 <- recode_meds_cycles3to6(cycle5, cycle5_meds, c("any_htn_med", "diab_med"))
cycle6 <- recode_meds_cycles3to6(cycle6, cycle6_meds, c("any_htn_med", "diab_med"))

cycle4_htn <- recode_after_meds(cycle4, htn_variables)
cycle5_htn <- recode_after_meds(cycle5, htn_variables)
cycle6_htn <- recode_after_meds(cycle6, htn_variables)

# Add cycle identifiers and combine
cycle3_htn$cycle <- 3
cycle4_htn$cycle <- 4
cycle5_htn$cycle <- 5
cycle6_htn$cycle <- 6

combined <- bind_rows(cycle3_htn, cycle4_htn, cycle5_htn, cycle6_htn)

# Hypertension prevalence by cycle
combined |>
  filter(!is.na(htn_status)) |>
  group_by(cycle) |>
  summarise(
    n = n(),
    n_htn = sum(htn_status == 1),
    prevalence = mean(htn_status == 1)
  )

Next steps

• Add more drug classes – This walkthrough used any_htn_med and diab_med, but chmsflow supports 8 drug classes. See Recoding medications for the full list and how to use individual classification functions.
• Include cycles 1–2 – Cycles 1–2 use a different medication format but the same downstream workflow. See Recoding medications for the cycles 1–2 pipeline.
• Understand the recoding rules – To inspect or customize how variables are harmonized, see Variable schema reference.
• Handle missing data carefully – The tagged_na("a") and tagged_na("b") values in results carry different meanings. See Missing data (tagged_na).
• Work at an RDC – To run this analysis on real CHMS data, see Using chmsflow at an RDC.
• Understand the methodology – For why harmonization is non-trivial and how chmsflow works under the hood, see Methodology.