Surgeons sometimes replace a missing ureter with a segment of small intestine (an ileal interposition), or reroute urine through an intestinal segment after bladder removal. Either way, the join between ureter and bowel can scar shut. This narrowing, called a uretero-ileal anastomotic stricture, occurs in 3% to 30% of cases and slowly destroys the kidney above it. Materials scientists have built stent coatings that release an anti-scarring drug over about 12 weeks. The only such coating in our file was tested in rabbits for just 4 weeks, and never in bowel-to-ureter joins. In human series, strictures show up months later, with mean recurrence at 5.4 months after endoscopic repair and 11.8 months after surgical repair. Nobody knows whether a 12-week drug reservoir covers the period when scar-forming risk is actually high, or whether it runs out just before that period begins.
We propose to measure that period directly from existing patient records. Using retrospective cohorts of ileal urinary reconstruction, we will estimate the time-varying risk of stricture, meaning the chance per month of a new stricture given none so far. We will correct for the fact that strictures are only found when someone orders an imaging test, and treat death as a separate outcome that prevents a stricture from ever being observed. We will also test whether the risk curve lines up better with the date of surgery or with the date the stent is pulled out. The result is two design numbers the coating field currently lacks: how long a drug reservoir must last to cover most stricture risk, and how long any preclinical or early clinical study must follow patients before a "no stricture" result means anything.
Core Claim
The claim is that the timing of stricture risk, rather than its overall rate, is the missing quantity linking the drug-delivery work to the surgical outcome work, and that it can be recovered from routine records if detection and competing death are modeled explicitly. Suppose $F(t)$ is the cumulative incidence of stricture by time $t$ after surgery, estimated with death as a competing event and with strictures known only to lie between two imaging dates. Suppose also that $\tau$ is a long reference horizon, such as 36 months. Then the share of eventual strictures detected inside the window a reservoir of duration $D$ covers is
$$C(D) = \frac{F(D)}{F(\tau)}.$$
Detection lags the onset of scarring, so $C(D)$ computed on detection times is a conservative lower bound on how much of the scarring process a reservoir of length $D$ overlaps. For the 12-week coating in Sustained drug-releasing hydrogel coatings for ureteral stents to prevent iatrogenic injury-induced ureteral stricture, this coating releases about 82% of its drug by week 12. If $C(3\text{ months})$ is small, that design is aimed at the wrong window for bowel-to-ureter joins. If it is large, then the design's 4-week animal test is the weak link, and the timing is not.
The second claim concerns which clock is right. If curves aligned to stent removal agree across patients with different stent durations better than curves aligned to surgery, then stent removal is when risk is unmasked. In that case the reservoir must outlast the stent, whether through a biodegradable depot left at the join or a coating whose release continues past removal. This matters because stents in the Mou cohort stayed in for 4.2 to 6.5 months on average, which is longer than the coating's release life.
Why This Has Not Been Done
Three pieces of work motivate the question, and none answers it. The coating paper Sustained drug-releasing hydrogel coatings for ureteral stents to prevent iatrogenic injury-induced ureteral stricture built a polyacrylamide hydrogel on polyurethane stents carrying pirfenidone-loaded nanoparticles. It showed reduced collagen deposition and swelling of the kidney from blocked urine (hydronephrosis) in injured rabbit ureters, but followed the animals for only 4 weeks and never modeled the bowel-join environment. It supplies the intervention whose timing we want to judge. We do not test the coating itself.
The therapeutic analysis of benign uretero-ileal anastomotic stricture after radical cystectomy and urinary diversion. followed 47 patients treated for stricture. It reports mean recurrence of 5.4 months after endoscopic repair and 11.8 months after surgical reimplantation, a drop in kidney filtration from 24.1 to 17.9 mL/min after endoscopic repair, and endoscopic success as low as 6% for strictures longer than 1 cm. It is the strongest evidence that failures are late. However, a mean time-to-recurrence among those who recurred ignores censored patients and says nothing about the shape of the risk curve.
The Asian consortium analysis Long-Term Outcomes of Orthotopic Neobladder Versus Ileal Conduit Urinary Diversion in Robot-Assisted Radical Cystectomy (RARC): Multicenter Results from the Asian RARC Consortium pooled 521 patients from nine centers. It showed with multivariable Cox regression that an apparent survival advantage for neobladders came from younger, healthier patients being selected for them (disease-free survival hazard ratio 0.824, p = 0.453). That shows the registry can support adjusted analyses. The same analysis, however, did not record stricture dates, stent removal dates, or suture technique, so it constrains us: the needed variables must be added by chart review.
The surgical-technique literature on tension-spreading U-shaped mattress sutures, such as the modified Blumgart approach in Laparoscopic hepaticopancreaticoduodenectomy for synchronous intrahepatic cholangiocarcinoma and distal common bile duct tumors: a case report, and intracorporeal ileal interposition in Robotic Bilateral Ileal Interposition: The Total Intracorporeal Approach, describes how joins are built. It contains no follow-up data on when they fail. We use suture configuration and approach only as covariates.
Research Program
The first step is data. We will extend the Asian consortium registry and the Zhejiang stricture cohort by chart abstraction. For each patient we will record every imaging date and its result (hydronephrosis present or absent), every stent insertion and removal date, serial creatinine-based filtration estimates, death dates, and the join technique where the operative note allows (for example, a U-shaped or mattress anchoring stitch, running barbed suture, or a standard separate-ureter join). Ileal interposition cases for ureteral loss from contributing centers will be pooled as a separate stratum. Those cases will be few, so they get descriptive curves rather than adjusted estimates.
The second step is estimation. A stricture is known only to have appeared between the last clear image and the first abnormal one. We will therefore fit a discrete-time competing-risks model in monthly intervals, using the PyDTS package (PyDTS: A Python Package for Discrete Time Survival Analysis with Competing Risks; code at tomer1812/pydts). The model has stricture and death as competing events and adjusts for age, performance status, baseline filtration, prior radiation, diversion type, and center. Interval censoring will be handled by distributing each detected event over its bracketing interval using the imaging schedule. The imaging schedule itself will be modeled as a function of covariates, so that sites that image more often do not appear to have earlier strictures. The model will be fit twice, once on the surgery clock and once on the stent-removal clock, with stent duration as a covariate in both.
The third step is translation. From the fitted curves we will report $C(D)$ for reservoir lengths from 4 weeks to 12 months. We will also report the follow-up length a study needs before a zero-event result excludes, say, half of eventual strictures. The same curves will be computed separately for the post-endoscopic-repair setting in the Zhejiang cohort, because that is where a coated stent would most plausibly be tried first. That cohort has only 19 endoscopic cases, so it yields wide uncertainty bands rather than point targets.
Baselines
The naive reference point is what the field reports today: a crude stricture proportion at whatever follow-up each center happens to have, plus a mean time-to-recurrence among patients who recurred. This is the summary Mou et al. give, with means of 5.4 and 11.8 months. We will compute it on exactly the same records. It is beaten if our estimates predict late-period stricture counts in held-out data better, since crude summaries ignore censoring and cannot extrapolate.
The strongest existing method is the analysis style of the consortium paper: a multivariable Cox model, here fitted to time-to-first-abnormal-image, together with Kaplan–Meier curves. That paper reports only survival hazard ratios, and no reported result was found on Lacuna for a stricture-specific Cox model in ileal reconstruction. We will fit it with the same covariates, the same data, and the same cross-validation folds as our model. Our model beats it if it gives better-calibrated cumulative incidence at 12, 24, and 36 months in temporal validation. A second test is whether the two approaches imply materially different $C(D)$, specifically a difference in $C(3\text{ months})$ larger than the bootstrap uncertainty. If they agree, the simpler model wins, and we will say so.
Three ablations isolate the key ingredients. The first drops interval censoring and uses detection dates directly. The second treats death as ordinary censoring, which inflates stricture incidence in an elderly cohort. The third uses only the surgery clock. Each ablation is judged by the same held-out calibration and by how much it shifts the reservoir target.
Why not just…?
Why not just run a longer animal study of the coating? A 6- or 12-month rabbit study is needed eventually, but it cannot say what horizon matters in humans. Rabbit ureters with focal burn injury heal on their own timescale and lack bowel mucus and bacteria. Without a human risk curve, the animal study has no target duration. Our output is what would set that study's length and endpoints.
Why not just report Kaplan–Meier stricture-free curves? This is the real threat. If death before stricture is rare and imaging schedules are similar across centers, a standard curve will give nearly the same $C(D)$. After bladder removal, however, deaths from cancer are common, and ignoring them overstates stricture risk. Detection dates also cluster at routine imaging visits, which shifts apparent timing. The ablations above measure exactly how large these distortions are. If they are small, we will recommend the simpler method.
Why not use deep survival clustering to find high-risk subgroups? Methods such as Deep Clustering Survival Machines with Interpretable Expert Distributions (LogRank 1067.6 on the SUPPORT dataset, versus 563.4 for Deep Survival Machines) and A Deep Variational Approach to Clustering Survival Data were built for cohorts with thousands of patients and many events. A few hundred reconstructions with perhaps tens of strictures cannot support them, and the question here is timing, not finding subgroups. We deliberately leave them out.
Why not just start a prospective trial of coated stents? A trial without a timing target risks choosing the wrong follow-up. Clinical complete response as a surrogate for pathological response in bladder cancer: a systematic review and meta‐analysis showed in a related bladder cancer setting that short-term clinical restaging agrees with the final tissue result only 51% of the time (95% CI 0.42–0.60). Early urology endpoints can mislead. Our study is cheap and fast, and it sets the trial's minimum follow-up.
Decisive Checks
Temporal validation is the main test. We fit on consortium cases from 2007 to 2015 and predict stricture counts by month for 2016 to 2020. This checks whether any model captures the late risk rather than just the early events.
Clock alignment is decided by comparing held-out likelihood and by testing whether stent duration still predicts timing on each clock. If stent duration stops predicting timing on the removal clock, risk is being unmasked at removal.
Surveillance bias is checked in a simulation using each center's real imaging intervals. We generate strictures with a known onset curve and confirm that our estimator recovers it while the naive estimate does not.
The headline result is $C(3\text{ months})$ with its uncertainty, for primary reconstruction and for post-endoscopic salvage. The decision rule is simple. A value well below one-half means current 12-week coatings are too short for this setting and depot design should target the post-removal window. A value near one means reservoir timing is not the bottleneck.
References
- Sustained drug-releasing hydrogel coatings for ureteral stents to prevent iatrogenic injury-induced ureteral stricture
- The therapeutic analysis of benign uretero-ileal anastomotic stricture after radical cystectomy and urinary diversion.
- Long-Term Outcomes of Orthotopic Neobladder Versus Ileal Conduit Urinary Diversion in Robot-Assisted Radical Cystectomy (RARC): Multicenter Results from the Asian RARC Consortium
- Laparoscopic hepaticopancreaticoduodenectomy for synchronous intrahepatic cholangiocarcinoma and distal common bile duct tumors: a case report
- Robotic Bilateral Ileal Interposition: The Total Intracorporeal Approach
- PyDTS: A Python Package for Discrete Time Survival Analysis with Competing Risks
- Deep Clustering Survival Machines with Interpretable Expert Distributions
- A Deep Variational Approach to Clustering Survival Data
- Clinical complete response as a surrogate for pathological response in bladder cancer: a systematic review and meta‐analysis