Table 2. Covariate-adjusted intervention effects for outcomes at 6-month follow-up and 18-month follow-up.
| Outcome | Period | Interventiona | Controla | Intervention-control difference at 6-/18-month follow-up versus difference at baseline (95% CI; P value)b | Intervention-control difference at 18-month follow-up versus difference at 6 months (95% CI; P value)b |
|---|---|---|---|---|---|
| APR☯ | |||||
| Baseline | 1,171/1,400 (84%) | 1,063/1,400 (76%) | |||
| 6 months | 515/1,380 (37%) | 1,084/1,400 (77%) | −49 pp (−63 to −35); <0.0001 | ||
| 18 months | 2,748/5,084 (54%) | 2,772/3,685 (75%) | −36 pp (−55 to −17); <0.0001 | 13 pp (−7 to 33); 0.21 | |
| Multiple APR¶ | |||||
| Baseline | 103/1,171 (9%) | 83/1,063 (8%) | |||
| 6 months | 29/515 (6%) | 65/1,084 (6%) | −2 pp (−7 to 3); 0.51 | ||
| 18 months | 54/2,748 (2%) | 209/2,772 (8%) | −6 pp (−13 to 0); 0.048 | −5 pp (−8 to −1); 0.008 | |
| Broad-spectrum APR☯ | |||||
| Baseline | 942/1,171 (80%) | 787/1,063 (74%) | |||
| 6 months | 346/515 (67%) | 794/1,084 (73%) | −12 pp (−21 to −4); 0.005 | ||
| 18 months | 2,089/2,748 (76%) | 2,082/2,772 (75%) | −20 pp (−34 to −6); 0.006 | −8 pp (−22 to 6); 0.28 | |
| Infusion APR‡ | |||||
| Baseline | 252/1,171 (22%) | 244/1,063 (23%) | |||
| 6 months | 110/515 (21%) | 331/1,084 (31%) | −8 pp (−23 to 7); 0.31 | ||
| 18 months | 266/2,748 (10%) | 365/2,772 (13%) | −2 pp (−15 to 11); 0.8 | 6 pp (−12 to 24); 0.5 | |
| Antiviral prescription rate☯ | |||||
| Baseline | 901/1,400 (64%) | 609/1,400 (44%) | |||
| 6 months | 942/1,380 (68%) | 732/1,400 (52%) | −3 pp (−18 to 11); 0.64 | ||
| 18 months | 3,017/5,084 (59%) | 1,778/3,685 (48%) | −3 pp (−29 to 24); 0.85 | 1 pp (−21 to 22); 0.93 | |
| Glucocorticoid prescription rate¶ | |||||
| Baseline | 326/1,400 (23%) | 304/1,400 (22%) | |||
| 6 months | 298/1,380 (22%) | 254/1,400 (18%) | 2 pp (−11 to 14); 0.8 | ||
| 18 months | 1,149/5,084 (23%) | 585/3,685 (16%) | 3 pp (−10 to 15); 0.69 | 1 pp (−11 to 13); 0.87 | |
| Vitamin prescription rate☯ | |||||
| Baseline | 192/1,400 (14%) | 245/1,400 (18%) | |||
| 6 months | 166/1,380 (12%) | 223/1,400 (16%) | 0 pp (−7 to 7); 0.97 | ||
| 18 months | 936/5,084 (18%) | 700/3,685 (19%) | 4 pp (−3 to 11); 0.24 | 4 pp (−5 to 13); 0.41 | |
| Traditional Chinese medicine prescription rate☯ | |||||
| Baseline | 1,152/1,400 (82%) | 999/1,400 (71%) | |||
| 6 months | 1,231/1,380 (89%) | 1096/1,400 (78%) | −1 pp (−13 to 12); 0.92 | ||
| 18 months | 4,501/5,084 (89%) | 3,077/3,685 (84%) | −2 pp (−12 to 7); 0.66 | −1 pp (−8 to 5); 0.66 | |
| Nonantibiotic medicine prescription rate☯ | |||||
| Baseline | 1,218/1,400 (87%) | 1,194/1,400 (85%) | |||
| 6 months | 1,179/1,380 (85%) | 1,215/1,400 (87%) | −2 pp (−8 to 4); 0.52 | ||
| 18 months | 4,428/5,084 (87%) | 3,255/3,685 (88%) | −6 pp (−16 to 3); 0.17 | −4 pp (−10 to 1); 0.1 | |
| Full prescription cost (USD)¶ | |||||
| Baseline | 4.2 (±1.6) | 4.4 (±1.9) | |||
| 6 months | 4.2 (±1.6) | 4.4 (±2.0) | −0.09 (−0.34 to 0.15); 0.46 | ||
| 18 months | 4.5 (±2.2) | 4.7 (±3.0) | 0.02 (−0.41 to 0.45); 0.93 | 0.11 (−0.36 to 0.58); 0.64 | |
| Antibiotics cost (USD)¶ | |||||
| Baseline | 0.6 (±0.4) | 0.5 (±0.4) | |||
| 6 months | 0.3 (±0.4) | 0.5 (±0.4) | −0.35 (−0.45 to −0.25); <0.0001 | ||
| 18 months | 0.4 (±0.4) | 0.5 (±0.4) | −0.26 (−0.38 to −0.13); <0.0001 | 0.09 (−0.05 to 0.23); 0.19 | |
| Other medication cost (USD)¶ | |||||
| Baseline | 2.2 (±1.5) | 2.4 (±1.9) | |||
| 6 months | 2.5 (±1.5) | 3.1 (±2.7) | 0.24 (0.01 to 0.48); 0.043 | ||
| 18 months | 2.7 (±2.1) | 2.8 (±2.9) | 0.24 (−0.16 to 0.64); 0.24 | 0.00 (−0.46 to 0.45); 0.99 |
aIntervention and control-arm summary data for prescribing rate outcomes are the number of prescriptions containing the relevant medicine divided by the total number of prescriptions (percentage), and for cost outcomes are mean (±SD).
bEstimated treatment effects represent either the difference between the intervention minus control difference at 6 months and the intervention minus control difference at baseline, or the difference between the treatment effect at 18 months and the treatment effect at baseline months, or they represent the difference between the treatment effect at 18 months and the treatment effect at 6 months as indicated, after adjusting for patient sex, age, and insurance payment status, and prescribing doctor sex, age, and education level.
Treatment effects for prescribing rate outcomes are on the absolute pp scale, and for cost outcomes are in USD per prescription, with each prescription representing one patient–doctor consultation. The between-time-period difference in treatment-arm differences (and the associated 95% CIs and P values) are estimated by GEE coefficients for the interaction between treatment arm and time period. The GEEs use either binomial errors and an identity link (☯, ‡) for binary prescribing outcomes, or Gaussian errors and an identity link (¶) for continuous cost outcomes or those binary prescribing outcomes where the binomial and then Poisson identity models failed to converge. GEEs accounted for clustering within facilities and within facilities across time periods either using an exchangeable correlation matrix (☯, ¶) or an identity matrix (‡) where the GEEs failed to converge with an exchangeable correlation matrix, given that GEEs are robust to misspecification of the correlation matrix. Outcome data were present for all outcomes and time periods, but <1% of covariate data were missing (see S3 Table). Analyses therefore excluded all patient prescriptions with missing covariate data and assume that data are missing at random. There were no changes to the original allocation of facilities.
Abbreviations: APR, antibiotic prescription rate; GEE, generalised estimating equation; pp, percentage points; USD, US dollars.