{
  "abstract": "Description Health systems need methods to evaluate quality improvement (QI) interventions that are quick, practical and analytically robust, even when outcome events are rare. We often must choose between two plausible options, and defaulting to the status quo because we haven’t met conventional significance thresholds (p <0.05) may not be the best strategy. Here we present one such example, used to assess whether post-discharge phone calls can be safely replaced with text messages - reducing staff workload without compromising care quality.470 patients discharged after elective outpatient surgery were randomized over 3 months to call or text (table 1). Outcomes were 14-day readmissions and emergency department (ED) visits, both of which occur at low rates. ED visits and readmissions did not differ significantly between groups, though power was limited (figure 1). To assess the likelihood of missed harm, 10,000 bootstrap simulations were conducted (figure 2). Text patients had worse outcomes in only 12% of simulations for ED visits and 31% for admissions (figures 3 and 4).In conclusion, a short, randomized test found no signal of harm when switching post-discharge calls to texts, suggesting texts are a safe, resource-efficient alternative in this population. This method is a useful strategy for evaluating operational interventions that balances efficiency with analytic rigor.Abstract 39 Figure 1Hospital admission and ED visit rates within 14 days post-discharge by outreach type. Bar chart comparing 14-day hospital admission and emergency department (ED) visit rates between patients who received a follow-up phone call (purple, n=240) and those who received a text message (green, n=230) after discharge. Admission rates were 5.8% in the call group and 4.8% in the text group. ED visit rates were 1.7% for the call group and 0.4% for the text group. Error bars represent 95% confidence intervals. Differences were not statistically significantAbstract 39 Figure 2Bootstrap sampling approach for analyzing rare outcomes. Illustration of the bootstrap sampling method used to assess the likelihood of rare post-discharge events. The original dataset was resampled with replacement 10,000 times to simulate the distribution of emergency department visits and hospitalizations across the text and call groups. This approach enabled estimation of the probability that the text group had worse outcomes, despite low event ratesAbstract 39 Figure 3Simulated difference in 14-day ED visit rates between text and call groups (bootstrap analysis). Histogram with overlaid density curve showing the distribution of 10,000 bootstrap simulations comparing emergency department (ED) visit rates between the call and text message groups. The x-axis reflects the simulated difference (Call ED rate – Text ED rate). The shaded curve illustrates the probability density across simulated differences. Values left of the dashed line indicate simulations where the ED rate was higher in the text group. In only 12% of simulations did the text group have a worse ED rate. Text follow-up was equal to or better than calls in 88% of casesAbstract 39 Figure 4Simulated difference in 14-day hospital encounter rates between text and call groups (bootstrap analysis). Histogram with overlaid density curve showing the distribution of 10,000 bootstrap simulations comparing 14-day hospital encounter rates between the call and text groups. The x-axis represents the simulated difference (Call hospitalization rate – Text hospitalization rate). The shaded curve reflects the probability density across simulated outcomes. Values to the left of the dashed line indicate simulations where the text group had a higher hospital encounter rate. In 31% of simulations, the text group had worse outcomes, while in 69% of simulations, the text group performed equally or better in terms of hospitalizationsAbstract 39 Table 1Baseline characteristics of patients in call and text groups with standardized mean differences (SMDs). This table summarizes patient characteristics across the call (N=240) and text (N=230) groups. Standardized mean differences (SMDs) are provided to assess covariate balance; an SMD >0.25 was considered indicative of meaningful imbalance. Language (SMD = 0.314) and insurance (SMD = 0.306) exceeded this threshold. Sensitivity analyses adjusting for these variables showed no meaningful change in results Variables Call (N=240) Text (N=230) SMD Sex Female 103 (42.9%) 78 (33.9%) 0.186 Mean Age (SD) 67.5 (12.1) 66.0 (14.0) 0.114 Language 0.314 Bengali 36 (15.0%) 38 (16.5%) English 127 (52.9%) 118 (51.3%) Spanish 46 (19.2%) 42 (18.3%) Race 0.135 Asian 68 (28.3%) 68 (29.6%) Black or African American 29 (12.1%) 22 (9.6%) Hispanic or Latino or Spanish 45 (18.8%) 42 (18.3%) Missing 36 (15.0%) 38 (16.5%) White 56 (23.3%) 56 (24.3%) Discharge Department 0.098 TH HCC 13 72 (30.0%) 71 (30.9%) TH KP 3 46 (19.2%) 48 (20.9%) TH KP 5 CARDIAC CATH 24 (10.0%) 17 (7.4%) TH KP 5 PRE POST 98 (40.8%) 94 (40.9%) Insurance 0.306 MEDICAID 26 (10.8%) 13 (5.7%) MEDICARE 39 (16.3%) 21 (9.1%) Private 175 (72.9%) 196 (85.2%)",
  "authors": [
    {
      "affiliations": [
        "NYU Grossman School of Medicine"
      ],
      "name": "Olivia Korostoff-Larsson"
    },
    {
      "affiliations": [
        "NYU Grossman School of Medicine"
      ],
      "name": "Jeremy Lu"
    },
    {
      "affiliations": [
        "NYU Grossman School of Medicine"
      ],
      "name": "Cathy Jian"
    }
  ],
  "title": "39 Balancing speed and rigor: a short, randomized test to optimize post-discharge communication",
  "uid": "15d2b790-69eb-5382-bfda-4d30c5077ff4"
}