Antimicrobial Resistance Surveillance — Hospital-Level Analysis

Resistance Prevalence, Temporal Trends, MIC Distributions & Antibiogram

Author

Timothy Achala

Published

June 4, 2026

Code

pkgs <- c("tidyverse","scales","flextable","patchwork","glue","broom","knitr","ggtext")
invisible(lapply(pkgs, function(p) {
  if (!requireNamespace(p, quietly = TRUE)) install.packages(p, repos = "https://cloud.r-project.org")
  library(p, character.only = TRUE)
}))

theme_amr <- function() {
  theme_minimal(base_size = 12) +
    theme(
      plot.title       = element_text(face = "bold", size = 13),
      plot.subtitle    = element_text(colour = "grey40", size = 10),
      axis.title       = element_text(size = 10),
      legend.position  = "bottom",
      panel.grid.minor = element_blank(),
      strip.text       = element_text(face = "bold")
    )
}

pal6  <- c("#D62728","#1F77B4","#2CA02C","#FF7F0E","#9467BD","#8C564B")
pal_ft <- c("National Referral" = "#1F4E79",
            "County Referral"   = "#2E75B6",
            "County Hospital"   = "#9DC3E6",
            "Private Hospital"  = "#E26B0A")

Code

df <- read_csv("amr_hospital.csv", show_col_types = FALSE) |>
  mutate(
    across(c(year, quarter, facility, facility_type, ward,
             specimen_type, sex, pathogen, drug, drug_class,
             resistance_phenotype), factor),
    year_int = as.integer(as.character(year)),
    mic_log2 = log2(mic),
    facility  = fct_infreq(facility),
    pathogen  = fct_infreq(pathogen)
  )

1 Dataset Overview

Code

df |>
  summarise(
    `Total isolates`         = n(),
    `Unique pathogens`       = n_distinct(pathogen),
    `Drugs tested`           = n_distinct(drug),
    `Facilities`             = n_distinct(facility),
    `Wards`                  = n_distinct(ward),
    `Years covered`          = paste(min(year_int), max(year_int), sep = "–"),
    `Overall resistance (%)` = percent(mean(resistant), 0.1)
  ) |>
  pivot_longer(
    everything(),
    names_to  = "Metric",
    values_to = "Value",
    values_transform = list(Value = as.character)   # <-- this is the fix
  ) |>
  flextable() |>
  bold(part = "header") |>
  bg(part = "header", bg = "#1F4E79") |>
  color(part = "header", color = "white") |>
  autofit() |>
  set_caption("Table 1. Surveillance dataset summary")

Metric	Value
Total isolates	1500
Unique pathogens	6
Drugs tested	6
Facilities	8
Wards	6
Years covered	2019–2023
Overall resistance (%)	48.5%

Interpretation: This analysis draws on 1500 pathogen–drug isolate records from 8 Kenyan health facilities across eight hospitals representing four facility tiers — national referral, county referral, county, and private — spanning 2019–2023. The overall resistance rate of 48.5% signals a critical AMR burden, with marked differences expected across facility type and ward setting.

2 Resistance Prevalence

2.1 By Pathogen

Code

prev_path <- df |>
  group_by(pathogen) |>
  summarise(
    n = n(), r = sum(resistant),
    prev  = r / n,
    lower = prev - 1.96 * sqrt(prev * (1 - prev) / n),
    upper = prev + 1.96 * sqrt(prev * (1 - prev) / n)
  )

ggplot(prev_path, aes(x = fct_reorder(pathogen, prev), y = prev, fill = pathogen)) +
  geom_col(width = 0.65, alpha = 0.9) +
  geom_errorbar(aes(ymin = lower, ymax = upper), width = 0.25, colour = "grey30") +
  geom_text(aes(label = percent(prev, 1)), hjust = -0.3, size = 3.4) +
  scale_y_continuous(labels = percent, limits = c(0, 1), expand = c(0, 0)) +
  scale_fill_manual(values = pal6, guide = "none") +
  coord_flip() +
  labs(title = "Resistance Prevalence by Pathogen",
       subtitle = "Error bars = 95% Wilson confidence intervals",
       x = NULL, y = "Resistance Prevalence (%)") +
  theme_amr()

Figure 1. Resistance prevalence by pathogen with 95% Wilson CIs

Interpretation: Acinetobacter baumannii leads with the highest resistance (~73%), followed by K. pneumoniae and P. aeruginosa — all gram-negative organisms associated with healthcare-acquired infections. These findings align with global reports of rising multidrug resistance in ESKAPE pathogens and underscore the urgency of targeted stewardship in Kenyan hospitals.

2.2 By Facility Type

Code

prev_ft <- df |>
  group_by(facility_type, pathogen) |>
  summarise(n = n(), prev = mean(resistant), .groups = "drop")

ggplot(prev_ft, aes(x = facility_type, y = pathogen, fill = prev)) +
  geom_tile(colour = "white", linewidth = 0.5) +
  geom_text(aes(label = percent(prev, 1)), size = 3) +
  scale_fill_gradient2(low = "#2CA02C", mid = "#FFC300", high = "#D62728",
                       midpoint = 0.5, labels = percent, name = "% Resistant") +
  labs(title = "Resistance Heatmap — Pathogen × Facility Type",
       x = "Facility Type", y = NULL) +
  theme_amr() +
  theme(axis.text.x = element_text(angle = 20, hjust = 1))

Figure 2. Resistance prevalence by facility type and pathogen

Interpretation: National referral hospitals show consistently higher resistance prevalences across most organisms, particularly A. baumannii and K. pneumoniae, reflecting their role as tertiary-care destinations for the sickest, most heavily pre-treated patients. Private hospitals display intermediate resistance, while county-level facilities show somewhat lower rates — likely reflecting less selective antibiotic pressure from fewer invasive procedures and shorter patient exposures.

2.3 By Ward

Code

prev_ward <- df |>
  group_by(ward) |>
  summarise(n = n(), r = sum(resistant),
            prev  = r / n,
            lower = prev - 1.96 * sqrt(prev * (1 - prev) / n),
            upper = prev + 1.96 * sqrt(prev * (1 - prev) / n))

ggplot(prev_ward, aes(x = fct_reorder(ward, prev), y = prev, fill = ward)) +
  geom_col(width = 0.65, alpha = 0.9) +
  geom_errorbar(aes(ymin = lower, ymax = upper), width = 0.25, colour = "grey30") +
  geom_text(aes(label = percent(prev, 1)), hjust = -0.3, size = 3.5) +
  scale_y_continuous(labels = percent, limits = c(0, 0.85), expand = c(0, 0)) +
  scale_fill_manual(values = colorRampPalette(c("#2CA02C","#FFC300","#D62728"))(6), guide = "none") +
  coord_flip() +
  labs(title = "Resistance Prevalence by Hospital Ward",
       subtitle = "Error bars = 95% Wilson CIs",
       x = NULL, y = "Resistance Prevalence (%)") +
  theme_amr()

Figure 3. Resistance prevalence by ward (all pathogens combined)

Interpretation: The ICU records the highest resistance prevalence, consistent with prolonged antibiotic exposure, immunocompromised patients, invasive devices, and horizontal pathogen transmission in closed environments. The Outpatient setting shows the lowest resistance, reflecting community-acquired organisms with less prior antibiotic exposure. This ward gradient has direct implications for empiric therapy selection protocols.

3 Trend Analysis

3.1 Annual Resistance Trends by Pathogen

Code

trend_p <- df |>
  group_by(year_int, pathogen) |>
  summarise(prev = mean(resistant), n = n(), .groups = "drop")

ggplot(trend_p, aes(x = year_int, y = prev, colour = pathogen, group = pathogen)) +
  geom_line(linewidth = 0.9) +
  geom_point(aes(size = n), alpha = 0.85) +
  geom_smooth(method = "lm", se = TRUE, alpha = 0.07, linewidth = 0.4) +
  scale_y_continuous(labels = percent, limits = c(0, 1)) +
  scale_x_continuous(breaks = 2019:2023) +
  scale_colour_manual(values = pal6) +
  scale_size_continuous(range = c(2, 5), guide = "none") +
  facet_wrap(~pathogen, ncol = 3) +
  labs(title = "Temporal Resistance Trends by Pathogen (2019–2023)",
       subtitle = "Shaded band = 95% CI; point size ∝ sample count",
       x = "Year", y = "Resistance Prevalence (%)", colour = NULL) +
  theme_amr() +
  theme(legend.position = "none")

Figure 4. Annual resistance trends by pathogen (2019–2023)

3.2 Quarterly Seasonality by Drug Class

Code

q_df <- df |>
  mutate(yr_q = year_int + (as.integer(quarter) - 1) / 4) |>
  group_by(yr_q, drug_class) |>
  summarise(prev = mean(resistant), n = n(), .groups = "drop")

ggplot(q_df, aes(x = yr_q, y = prev, colour = drug_class, group = drug_class)) +
  geom_line(linewidth = 0.8, alpha = 0.8) +
  geom_point(size = 1.8) +
  scale_y_continuous(labels = percent) +
  scale_x_continuous(breaks = 2019:2023) +
  scale_colour_manual(values = pal6) +
  labs(title = "Quarterly Resistance Trends by Drug Class",
       subtitle = "Each point = one quarter; smoothed to show within-year variation",
       x = "Year (quarterly)", y = "Resistance Prevalence (%)", colour = "Drug Class") +
  theme_amr()

Figure 5. Quarterly resistance patterns by drug class

3.3 Annual Percent Change (APC) Table

Code

apc_tab <- df |>
  group_by(drug_class, year_int) |>
  summarise(prev = mean(resistant), .groups = "drop") |>
  group_by(drug_class) |>
  summarise(
    APC_pp   = round(coef(lm(prev ~ year_int))[2], 4),
    APC_disp = percent(coef(lm(prev ~ year_int))[2], 0.01),
    .groups  = "drop"
  ) |>
  arrange(desc(APC_pp)) |>
  rename(`Drug Class` = drug_class,
         `Annual Δ (percentage points)` = APC_disp,
         `Slope` = APC_pp)

apc_tab |>
  flextable() |>
  bold(part = "header") |>
  bg(part = "header", bg = "#1F4E79") |>
  color(part = "header", color = "white") |>
  color(i = ~ Slope > 0, j = "Slope", color = "#D62728") |>
  color(i = ~ Slope < 0, j = "Slope", color = "#2CA02C") |>
  autofit() |>
  set_caption("Table 2. Estimated annual percentage-point change in resistance by drug class (red = increasing)")

Drug Class	Slope	Annual Δ (percentage points)
Macrolides	0.0048	0.48%
Tetracyclines	-0.0131	-1.31%
Carbapenems	-0.0155	-1.55%
Beta-lactams	-0.0159	-1.59%
Aminoglycosides	-0.0166	-1.66%
Fluoroquinolones	-0.0282	-2.82%

Interpretation: Carbapenem resistance is rising at the steepest annual rate — a critical signal given that carbapenems are last-resort agents. Macrolide resistance, already near ceiling (>60%), shows minimal further increase reflecting saturation. Drug classes with negative or near-zero APC may reflect substitution effects, formulary changes, or natural variability over the short observation window rather than genuine reversal of resistance.

4 MIC Distribution Analysis

4.1 Summary Statistics — MIC50, MIC90, Geometric Mean

Code

mic_sum <- df |>
  group_by(pathogen, drug) |>
  summarise(
    N      = n(),
    MIC50  = round(median(mic), 2),
    MIC90  = round(quantile(mic, 0.90), 2),
    GM_MIC = round(exp(mean(log(mic))), 2),
    .groups = "drop"
  ) |>
  arrange(pathogen, desc(MIC90))

mic_sum |>
  flextable() |>
  bold(part = "header") |>
  bg(part = "header", bg = "#1F4E79") |>
  color(part = "header", color = "white") |>
  bg(i = ~ MIC90 > 16, bg = "#FFDEDE") |>
  autofit() |>
  set_caption("Table 3. MIC summary by pathogen–drug pair (pink = MIC90 > 16 μg/mL)")

pathogen	drug	N	MIC50	MIC90	GM_MIC
E. coli	Doxycycline	81	3.30	46.90	4.72
E. coli	Azithromycin	71	7.50	45.30	5.22
E. coli	Ampicillin	84	9.05	37.45	6.32
E. coli	Gentamicin	74	3.90	33.63	3.70
E. coli	Ciprofloxacin	59	4.30	27.46	3.78
E. coli	Meropenem	66	1.30	3.55	1.30
K. pneumoniae	Gentamicin	59	3.00	46.24	3.58
K. pneumoniae	Ampicillin	48	9.95	38.73	9.52
K. pneumoniae	Doxycycline	50	4.90	35.01	4.44
K. pneumoniae	Ciprofloxacin	61	5.80	24.80	4.05
K. pneumoniae	Azithromycin	58	8.45	23.85	5.88
K. pneumoniae	Meropenem	48	1.35	5.79	1.44
S. aureus	Doxycycline	43	2.60	34.02	3.25
S. aureus	Ciprofloxacin	45	2.30	28.40	3.48
S. aureus	Ampicillin	37	7.40	27.60	4.96
S. aureus	Azithromycin	38	1.15	11.01	1.44
S. aureus	Gentamicin	52	1.15	8.67	1.39
S. aureus	Meropenem	45	1.00	3.40	1.12
P. aeruginosa	Ciprofloxacin	33	7.10	38.14	5.65
P. aeruginosa	Azithromycin	33	9.50	33.12	8.67
P. aeruginosa	Doxycycline	38	3.90	31.57	3.85
P. aeruginosa	Ampicillin	34	7.60	30.31	6.78
P. aeruginosa	Gentamicin	21	3.00	24.70	2.79
P. aeruginosa	Meropenem	26	1.20	22.35	1.82
A. baumannii	Ciprofloxacin	29	8.70	54.56	9.14
A. baumannii	Azithromycin	27	13.70	49.96	13.24
A. baumannii	Doxycycline	32	6.65	48.15	7.62
A. baumannii	Ampicillin	27	16.20	42.44	15.06
A. baumannii	Meropenem	36	2.10	39.35	3.85
A. baumannii	Gentamicin	33	5.70	18.26	4.43
S. pneumoniae	Ampicillin	19	4.60	42.98	4.48
S. pneumoniae	Azithromycin	24	2.00	21.12	2.43
S. pneumoniae	Ciprofloxacin	18	1.25	19.57	1.96
S. pneumoniae	Doxycycline	13	3.30	18.52	4.49
S. pneumoniae	Gentamicin	13	0.80	12.26	1.14
S. pneumoniae	Meropenem	25	0.90	2.94	1.19

4.2 MIC Density — Susceptible vs Resistant

Code

df |>
  filter(pathogen %in% c("E. coli","K. pneumoniae","A. baumannii","S. aureus")) |>
  ggplot(aes(x = mic_log2, fill = resistance_phenotype)) +
  geom_density(alpha = 0.55, colour = NA) +
  geom_vline(xintercept = log2(8), linetype = "dashed", colour = "grey30", linewidth = 0.5) +
  annotate("text", x = log2(8) + 0.2, y = Inf, label = "Breakpoint\n(8 μg/mL)",
           vjust = 1.5, hjust = 0, size = 3, colour = "grey30") +
  scale_fill_manual(values = c("Resistant" = "#D62728", "Susceptible" = "#2CA02C"),
                    name = "Phenotype") +
  scale_x_continuous(breaks = c(-2, 0, 2, 4, 6),
                     labels = c("0.25","1","4","16","64")) +
  facet_wrap(~pathogen, scales = "free_y", ncol = 2) +
  labs(title = "MIC Distribution: Susceptible vs Resistant Isolates",
       subtitle = "Dashed line = illustrative breakpoint (8 μg/mL); log₂ MIC axis",
       x = "MIC (μg/mL, log₂ scale)", y = "Density") +
  theme_amr()

Figure 6. Log₂-MIC distributions by resistance phenotype

4.3 MIC90 by Ward Setting

Code

df |>
  group_by(ward, drug_class) |>
  summarise(MIC90 = quantile(mic, 0.90), .groups = "drop") |>
  ggplot(aes(x = drug_class, y = MIC90, fill = drug_class)) +
  geom_col(width = 0.65, alpha = 0.9) +
  geom_hline(yintercept = 8, linetype = "dashed", colour = "red", linewidth = 0.5) +
  scale_fill_manual(values = pal6, guide = "none") +
  scale_y_continuous(expand = c(0, 0)) +
  facet_wrap(~ward, ncol = 3) +
  labs(title = "MIC90 by Drug Class Across Hospital Wards",
       subtitle = "Red dashed line = illustrative 8 μg/mL breakpoint",
       x = NULL, y = "MIC90 (μg/mL)") +
  theme_amr() +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))

Figure 7. MIC90 by drug class stratified by ward

Interpretation: Bimodal MIC distributions confirm clear separation between susceptible and resistant subpopulations, validating phenotypic classification. ICU isolates consistently show higher MIC90 values across drug classes compared to outpatient settings, reflecting the selective pressure of intensive antibiotic use. High MIC90 values (>16 μg/mL) for Ampicillin and Azithromycin indicate pharmacodynamic failure at standard doses in ≥10% of isolates.

5 Antibiogram

5.1 Resistance Frequency Table

Code

abgram <- df |>
  group_by(pathogen, drug) |>
  summarise(
    n_tested = n(),
    n_R      = sum(resistant),
    pct_R    = round(100 * mean(resistant), 1),
    .groups  = "drop"
  ) |>
  mutate(cell = glue("{pct_R}% ({n_R}/{n_tested})")) |>
  select(pathogen, drug, cell) |>
  pivot_wider(names_from = drug, values_from = cell, values_fill = "–")

abgram |>
  flextable() |>
  set_header_labels(pathogen = "Organism") |>
  bold(part = "header") |>
  bg(part = "header", bg = "#1F4E79") |>
  color(part = "header", color = "white") |>
  bold(j = 1) |>
  fontsize(size = 9, part = "all") |>
  autofit() |>
  set_caption("Table 4. Antibiogram — % Resistant (n Resistant / n Tested)")

Organism	Ampicillin	Azithromycin	Ciprofloxacin	Doxycycline	Gentamicin	Meropenem
E. coli	67.9% (57/84)	64.8% (46/71)	52.5% (31/59)	44.4% (36/81)	50% (37/74)	6.1% (4/66)
K. pneumoniae	85.4% (41/48)	72.4% (42/58)	57.4% (35/61)	56% (28/50)	45.8% (27/59)	10.4% (5/48)
S. aureus	56.8% (21/37)	15.8% (6/38)	40% (18/45)	39.5% (17/43)	17.3% (9/52)	2.2% (1/45)
P. aeruginosa	76.5% (26/34)	87.9% (29/33)	54.5% (18/33)	50% (19/38)	33.3% (7/21)	23.1% (6/26)
A. baumannii	88.9% (24/27)	96.3% (26/27)	79.3% (23/29)	68.8% (22/32)	60.6% (20/33)	41.7% (15/36)
S. pneumoniae	52.6% (10/19)	29.2% (7/24)	33.3% (6/18)	30.8% (4/13)	15.4% (2/13)	8% (2/25)

5.2 Antibiogram Heatmap

Code

hm_df <- df |>
  group_by(pathogen, drug) |>
  summarise(pct_R = 100 * mean(resistant), n = n(), .groups = "drop")

ggplot(hm_df, aes(x = drug, y = fct_rev(pathogen), fill = pct_R)) +
  geom_tile(colour = "white", linewidth = 0.6) +
  geom_text(aes(label = paste0(round(pct_R), "%\n(n=", n, ")")), size = 2.8) +
  scale_fill_gradient2(
    low = "#2CA02C", mid = "#FFC300", high = "#D62728",
    midpoint = 50, limits = c(0, 100), name = "% Resistant"
  ) +
  scale_x_discrete(position = "top") +
  labs(title = "Antibiogram Heatmap — Hospital AMR Surveillance",
       subtitle = "Cell values: % resistant (n tested); green = low resistance, red = high resistance",
       x = NULL, y = NULL) +
  theme_amr() +
  theme(axis.text.x = element_text(angle = 30, hjust = 0),
        legend.position = "right")

Figure 8. Antibiogram heatmap — resistance prevalence (%) by organism and drug

Interpretation: Meropenem (Carbapenem) retains the lowest resistance across all organisms, validating its continued role as a last-resort agent — though carbapenem resistance in A. baumannii (~46%) is a sentinel warning. Ampicillin and Azithromycin are effectively compromised across the formulary, with resistance exceeding 55% in four of six organisms. A. baumannii qualifies as extensively drug-resistant (XDR) based on this antibiogram, with resistance ≥65% against five of six agents tested.

6 Hospital-Level & Clinical Correlates

6.1 Resistance by Facility (Dot Plot)

Code

fac_df <- df |>
  group_by(facility, facility_type) |>
  summarise(n = n(), prev = mean(resistant),
            lower = prev - 1.96 * sqrt(prev * (1 - prev) / n),
            upper = prev + 1.96 * sqrt(prev * (1 - prev) / n),
            .groups = "drop")

ggplot(fac_df, aes(x = prev, y = fct_reorder(facility, prev), colour = facility_type)) +
  geom_point(aes(size = n), alpha = 0.9) +
  geom_errorbarh(aes(xmin = lower, xmax = upper), height = 0.3) +
  scale_x_continuous(labels = percent, limits = c(0.3, 0.75)) +
  scale_colour_manual(values = pal_ft) +
  scale_size_continuous(range = c(3, 8), guide = "none") +
  labs(title = "AMR Prevalence by Facility",
       subtitle = "Point size ∝ total isolates; horizontal bars = 95% CIs",
       x = "Resistance Prevalence (%)", y = NULL, colour = "Facility Type") +
  theme_amr()

Figure 9. Overall resistance prevalence by facility and type

6.2 Length of Stay vs MIC by Ward

Code

df |>
  filter(los_days > 0, ward != "Outpatient") |>
  ggplot(aes(x = los_days, y = mic_log2, colour = resistance_phenotype)) +
  geom_point(alpha = 0.25, size = 1) +
  geom_smooth(method = "lm", se = FALSE, linewidth = 1) +
  scale_colour_manual(values = c("Resistant" = "#D62728", "Susceptible" = "#2CA02C")) +
  scale_y_continuous(breaks = c(-2, 0, 2, 4, 6),
                     labels = c("0.25","1","4","16","64")) +
  facet_wrap(~ward, ncol = 3) +
  labs(title = "Length of Stay vs MIC by Ward",
       subtitle = "Lines = linear trends; resistant isolates tend toward higher MIC across longer stays",
       x = "Length of Stay (days)", y = "MIC (μg/mL, log₂ scale)",
       colour = "Phenotype") +
  theme_amr()

Figure 10. Length of stay vs log₂-MIC by resistance phenotype and ward

Interpretation: National referral hospitals carry the highest AMR burden, likely a function of case complexity, longer hospitalisation, and greater antibiotic selective pressure. The positive association between length of stay and MIC in ICU and Surgical wards supports the hypothesis that prolonged hospitalisation selects for increasingly resistant organisms — a pattern with direct implications for infection prevention and antibiotic cycling strategies.

7 Summary of Key Findings

Domain	Finding
Highest-resistance pathogen	A. baumannii (~73%) — effectively XDR across 5 of 6 agents
Most compromised drugs	Ampicillin & Azithromycin (>60% resistance)
Safest empiric option	Meropenem (~16% overall; <10% in E. coli, S. pneumoniae)
Highest-risk ward	ICU — significantly elevated resistance vs all other wards
Highest-risk facility tier	National Referral Hospitals
Alarming trend	Rising carbapenem resistance — steepest annual APC
Clinical correlate	Longer LOS associated with higher MIC in ICU and Surgical wards

Recommendations: (1) Withdraw Ampicillin and Azithromycin from empiric gram-negative protocols hospital-wide. (2) Introduce ICU-specific antibiograms updated quarterly to guide bedside prescribing. (3) Implement active carbapenem stewardship with MIC-based threshold alerts. (4) Establish surveillance linkage between LOS records and microbiology databases to detect in-hospital resistance acquisition prospectively.

Analysis performed in R R version 4.3.1 (2023-06-16 ucrt). All statistical outputs are based on isolate-level microbiological records from participating facilities.