Using the receiver operating characteristic curve to select pretreatment and pathologic predictors for early and late postprostatectomy PSA failure.

OBJECTIVES: Pretreatment prostate-specific antigen (PSA), prostatectomy Gleason score, margin status, and pathologic T stage are known explanatory variables for the postprostatectomy PSA outcome. We used the receiver operating characteristic (ROC) curve to select those factors that were optimal for predicting early and late postoperative PSA failure. METHODS: We designed and implemented a clinical outcome prediction expert that performs, assesses, and optimizes the actuarial prediction on individual cases. A postprostatectomy database of 1022 patients was divided into 60% for training and 40% for validation. The ROC areas of the predictors were calculated over a range of cutoff time from 24 to 60 months. RESULTS: Multivariate pathologic T stage/prostatectomy Gleason score/margin status had the highest ROC area of 0.900. Patients with Stage T disease less than T3, negative surgical margins, and Gleason score of 6 or less had a 90% probability to be PSA failure free at 4 years versus 36% otherwise. The pathologic T stage/margin status accurately predicted PSA failure at 24 months or less after prostatectomy with an ROC area of 0.800. Lower risk patients (less than Stage T3, negative surgical margins) had a 94% probability to be PSA failure free at 2 years versus 46% otherwise. CONCLUSIONS: A combination of actuarial analysis and ROC optimization accurately identified the individual patients at high risk of early and late postprostatectomy PSA failure.

ROC optimization may improve risk stratification of prostate cancer patients.

OBJECTIVES: Rational treatment decision requires accurate projection of the clinical course of a patient. Current methods in clinical outcome analysis mostly focus on population data. We investigated the applicability and optimization of the widely used actuarial method to project individual clinical outcomes. METHODS: We designed and implemented a Clinical Outcome Prediction Expert (COPE) that performs, assesses, and optimizes actuarial prediction on individual cases. We analyzed a post-prostatectomy database, consisting of 1043 patients. Sixty percent of the database was used for training and 40% for validation. Stratified actuarial curves are used to project individual outcomes. The prostate-specific antigen (PSA) level, the Gleason score, and the clinical American Joint Commission on Cancer Staging T-stage before treatment were used as predictors. The area under the receiver operator characteristic (ROC) curve was used to measure predictive performance. RESULTS: We obtained simple optimized stratification of pretreatment PSA level of 10 ng/mL or less, or more than 10 ng/mL; Gleason score of 6 or lower, or higher than 6; and clinical AJCC T-stage of T2a or lower, or higher. The optimized univariate risk scores were used to generate a multivariate score. After optimization, we found the higher risk group consisted of patients with PSA more than 10 ng/mL, or with PSA of 10 ng/mL or less and Gleason score higher than 6 and clinical AJCC T-stage higher than T2a. The optimized multivariate risk score has the highest ROC area of 0.77 among all predictors. CONCLUSIONS: The best conditions to perform actuarial prediction on individual cases are not known a priori and require optimization. This study shows that ROC optimization simplifies risk stratification and may improve the accuracy of clinical outcome prediction.

Beam spoilers versus bolus for 6 MV photon treatment of head and neck cancers.

When parallel opposed 6-MV x-ray beams are used for treatment of head and neck tumors, superficial tissues and lymphatics at shallow depths of < or =4 to 6 mm may be at cancer risk but receive less than full radiation dose. In these cases, the use of either a beam spoiler or bolus material can increase dose to superficial tissues. The potential benefit of a beam spoiler relative to bolus is preservation of skin-sparing characteristics for cases in which the skin surface does not require full dose. In this study, we evaluate the application of a beam spoiler and compare it to bolus for standard treatments of head and neck tumors. Measurements of both depth dose in-water and in-air profiles were made with a beam spoiler for a 6-MV photon beam. The measurements were combined with Monte Carlo calculations to obtain the energy spectrum of the spoiler-generated electrons. An in-house pencil beam treatment-planning algorithm was used to calculate the dose distribution with spoiler. The dose distribution in the buildup region was then studied with and without the spoiler for a typical head and neck treatment with parallel-opposed beams. Dose distributions and partial-volume dose histograms (PVDH) demonstrate the benefits provided by spoilers for the head and neck treatments and the limitations of their use. The beam spoiler is useful in treating the superficial lymphatics in the buildup region near head and neck tumors. Guidelines for use of beam spoiler versus bolus are discussed.

A quality assurance device for the accuracy of the isocentres of teletherapy and simulation machines.

A new quality assurance device has been designed to measure the location and wobble of the radiation isocentre of linacs and simulation machines as a function of gantry rotation. The radiation isocentre is the intersection in space of the central x rays of a linac or simulation machine at different gantry angles. Six radio-opaque markers 1 mm in size are embedded in a radio-transparent calibration object (specifically, a hollowed cube) in such a way that the markers are non-coplanar and uniquely identifiable in radiographic projections. The projective radiographs are obtained on the films held by the film holder attached to the gantry during the QA procedure. The marker positions of the calibration object define a known 3D reference frame, and their image positions on each radiograph determine the projective (3D to 2D matrix) transformation for that radiograph. Once the transformation is found, a 3D ray from the radiation source to any radiograph pixel becomes known. The radiographic pixels are coordinated (positioned and scaled) with respect to the projected image of radio-opaque fiducial cross-hairs fixed to a block tray and thus to the gantry. We select the central ray to correspond to the radiographic pixel whose rays at different gantry angles intersect in the smallest spatial domain. That pixel is found by a spiral search in the radiograph outward from the image of the radio-opaque cross-hair intersection. The wobble of the isocentre is defined by the set of points (on the central rays) at closest approach to the isocentre. The device was tested and compared with commercially available QA devices. It is able to locate the isocentre to within 0.5 mm. The offset of this derived radiation isocentre from the intersection of the positioning lasers can be found. To do this, the calibration object is initially placed so that the laser intersection point falls on a seventh radio-opaque marker near the centre of the hollow cube calibration object. The seventh marker is embedded in a thin radio-transparent rod that diagonally spans the hollowed space.

Determining clinical photon beam spectra from measured depth dose with the Cimmino algorithm.

A method to determine the spectrum of a clinical photon beam from measured depth-dose data is described. At shallow depths, where the range of Compton-generated electrons increases rapidly with photon energy, the depth dose provides the information to discriminate the spectral contributions. To minimize the influence of contaminating electrons, small (6 x 6 cm2) fields were used. The measured depth dose is represented as a linear combination of basis functions, namely the depth doses of monoenergetic photon beams derived by Monte Carlo simulations. The weights of the basis functions were obtained with the Cimmino feasibility algorithm, which examines in each iteration the discrepancy between predicted and measured depth dose. For 6 and 15 MV photon beams of a clinical accelerator, the depth dose obtained from the derived spectral weights was within about 1% of the measured depth dose at all depths. Because the problem is ill conditioned, solutions for the spectrum can fluctuate with energy. Physically realistic smooth spectra for these photon beams appeared when a small margin (about +/- 1%) was attributed to the measured depth dose. The maximum energy of both derived spectra agreed with the measured energy of the electrons striking the target to within 1 MeV. The use of a feasibility method on minimally relaxed constraints provides realistic spectra quickly and interactively.

Quality assurance of central axis dose data for photon beams by means of a functional representation of the tissue phantom ratio.

A recently proposed four-parameter functional representation for the tissue phantom ratio (TPR) in the domain of electronic equilibrium was tested for accuracy and applied to quality assurance of central axis dose data. The four parameters are energy dependent and are found for each beam by a minmax search. For photon energies of 4 MV and greater, the functional representation was accurate to within 1% of measured data for all depths and field sizes exceeding 10 cm. The representation was also found to be robust. With only nine measurements of the TPR (at the extremes and middle of the electronic equilibrium domain) used to determine the four parameters, the representation reproduced the TPR value of any depth and field size to within 1% of measured data for photon energies of 6 MV or greater. The representation was insensitive to random measurement errors of a magnitude comparable to that expected in clinical practice. Other findings indicated that the representation degrades gracefully as the domain is extended into regions not in electronic equilibrium. When used with a QA program, the functional representation of TPR provides a means of detecting and correcting errors of measurement and data transcription of central axis dose data.

Study to determine the ability of clinical laboratories to detect antimicrobial-resistant Enterococcus spp. in Buenos Aires, Argentina.

Few reports of vancomycin-resistant enterococci have appeared outside the USA. Therefore, we evaluated the ability of five laboratories in Buenos Aires, Argentina, to perform susceptibility testing using the disk diffusion method. Laboratories had difficulty identifying the low- and intermediate-level vancomycin-resistant phenotypes. This suggests that the disk diffusion method used by laboratories abroad may fail to detect some vancomycin-resistant enterococci.

Optimized lens-sparing treatment of retinoblastoma with electron beams.

PURPOSE: The ideal lens-sparing radiotherapy technique for retinoblastoma calls for 100% dose to the entire retina including the ora serrata and zero dose to the lens. Published techniques, most of which use photons, have not accomplished this ideal treatment. We describe here a technique that approaches this ideal configuration using electron beam therapy. METHODS AND MATERIALS: Dose-modeling calculations were made using a computer program built around a proprietary algorithm. This program calculates 3D dose distribution for electrons and photons and uses the Cimmino feasibility method for the inverse problem of beam weighting to achieve the prescribed dose. The algorithm has been verified in the ocular region by measurements in a RANDO phantom. To search for an ideal lens-sparing beam setup, a stylized phantom of an 8-month-old infant was generated with built-in inhomogeneities, and a phantom of a 5-year-old child was generated from a patient CT series. RESULTS: Of more than 100 different beam setups tested, two 9 MeV electron beams at gantry angles plus and minus 26 degrees from the optic nerve axis achieved the best distribution. Both fields have a lens block and an isocenter between the globe and origin of the optic nerve. When equal doses are given to both fields, the entire extent of the retina (including ora serrata) received 100%, while the lens received 10% or less. CONCLUSION: The two-oblique-electron-beam technique here described appears to meet most of the stringent dosimetry needed to treat retinoblastoma. It is suitable for a range of ages, from infancy to early childhood years.

Automated matching of corresponding seed images of three simulator radiographs to allow 3D triangulation of implanted seeds.

To match corresponding seed images in different radiographs so that the 3D seed locations can be triangulated automatically and without ambiguity requires (at least) three radiographs taken from different perspectives, and an algorithm that finds the proper permutations of the seed-image indices. Matching corresponding images in only two radiographs introduces inherent ambiguities which can be resolved only with the use of non-positional information obtained with intensive human effort. Matching images in three or more radiographs is an 'NP (Non-determinant in Polynomial time)-complete' problem. Although the matching problem is fundamental, current methods for three-radiograph seed-image matching use 'local' (seed-by-seed) methods that may lead to incorrect matchings. We describe a permutation-sampling method which not only gives good 'global' (full permutation) matches for the NP-complete three-radiograph seed-matching problem, but also determines the reliability of the radiographic data themselves, namely, whether the patient moved in the interval between radiographic perspectives.

Detection of cleavage products from an in vivo transcribed cis hairpin ribozyme in turnips using the CaMV plant virus.

In order to examine ribozyme (Rz) activity in vivo, we have adapted a virus to deliver Rz to plants. DNA fragments that code for both active and mutant cis-hairpin Rz were cloned into the double-stranded DNA plant virus, cauliflower mosaic virus (CaMV). These Rz constructs successfully infected Brassica campestris rapa (turnip). The plants that were infected with the active-Rz construct showed, on average, a one-week delay in the appearance of viral symptoms, when compared to the mutant-Rz control. Since CaMV replicates through reverse transcription of a full-length RNA intermediate, Rz cloned into the CaMV DNA should be transcribed within this viral RNA. If these Rz constructs cleave, the amount of intact virus RNA should be reduced, resulting in attenuated viral symptoms. In addition, lysate RNase protection assays showed fragments corresponding to the sizes of both the 5' and 3' cis cleavage products in the active Rz tissue. No cleavage products were observed from plant tissue infected with the mutant Rz. Both the attenuated systemic viral symptoms and the cleavage products from the protection assay strongly support in vivo transcription and cleavage of this hairpin Rz. This is the first report of an in vivo transcribed Rz showing cleaved products by direct RNA analysis (non-PCR) in plants or animals.

Calculation of the dose in the build-up region for high energy photon beam. Treatment planning when beam spoilers are employed.

Recoil electrons from a polystyrene sheet (beam spoiler) placed in a photon beam were used to modulate the dose in the build-up region. The effects of the beam spoiler on the dosimetry in simple phantoms were studied for both 6 MV and 15 MV photon beams at 100 cm source-to-surface distance (SSD) as a function of (1) the thickness of the spoiler, (2) the spoiler-to-surface distance (SD) and (3) the field size. A radiotherapy treatment planning (RTP) algorithm was used that employs pencil beam dose deposition kernels to evaluate the three-dimensional (3D) dose distribution due to photons and electrons in the build-up region. To determine the energy deposition kernels for the recoil electrons emanating from the spoiler, their energy spectrum needs to be known. Monte Carlo simulations were performed to derive the spectrum of electrons from the beam spoiler. The recoil electron spectrum was found to be nearly independent of field size, SD, and spoiler thickness. The pencil beam energy deposition kernel was calculated for the recoil electrons from the spoiler and inserted in the treatment planning algorithm. The dose calculation algorithm permits merging the photon and recoil electron dose components. In all cases the calculation agrees to within 3% of the measured results.

Alteration of hairpin ribozyme specificity utilizing PCR.

We have developed a method by which a researcher can quickly alter the specificity of a trans hairpin ribozyme. Utilizing this PCR method, two oligonucleotides, and any target vector, new ribozyme template sequences can be generated without the synthesis of longer oligonucleotides. We have produced templates with altered specificity for both standard and modified (larger) ribozymes. After transcription, these ribozymes show specific cleavage activity with the new substrate beta-glucuronidase (GUS), and no activity against the original substrate (HIV-1, 5' leader sequence). Utilizing this technique, it is also possible to produce an inactive ribozyme that can be used as an antisense control. Applications of this procedure would provide a rapid and economical system for the assessment of trans ribozyme activity.

Influence of cone design on the electron beam characteristics on clinical accelerators.

Electron beam characteristics were investigated for different applicator cone designs. Low-energy scattered electrons from the cone housing can contribute to the skin dose. The depth dose was measured as a function of electron energies and field size for different cones. The measurements were compared with calculated 3-D dose distributions obtained from Monte-Carlo-derived energy deposition kernels [Phys. Med. Biol. 37, 391 (1992)]. This comparison shows that beam-defining cones can be fabricated to eliminate low-energy scattered electrons from the cone housing. Eliminating low-energy electrons from the cones provides depth dose (DD) characteristics for clinical electron beams that are comparable to those (1) from nearly monoenergetic electrons (as determined from the practical range of the electrons), and (2) for which the DD is nearly independent of source-to-skin distance. Measurements of DD for small circular fields were also performed. For very small field sizes, corresponding to the width of the lateral spread of a pencil beam, the theoretical and measured DD were found to depend on field size. Three-dimensional dose calculations of small and irregularly shaped electron fields are necessary to appreciate these effects.

Semiautomated matching and seed position location for implanted ribbons.

A difficult, error-prone, and time consuming problem in brachytherapy is the determination of the three-dimensional (3-D) positions of implanted seeds. To triangulate 3-D seed positions, the corresponding seed images in radiographs taken at different parallax (gantry) angles must be identified (matched). With only two radiographs, there is an inherent problem of ambiguity in matching individual seed images. The use of the assignment algorithm cannot bypass this ambiguity. However, when seeds are implanted within ribbons, their positions in 3-D space can be determined accurately and (almost always) without ambiguity. This follows because the matching of ribbon images is much less ambiguous than the matching of individual seed images. We describe a semi-automated procedure to match the ribbon images on two radiographs, determine the 3-D loci of the ribbons, and locate the 3-D position of each seed on each ribbon. The ribbon matching problem involves a double application of the assignment algorithm. After image matching, the 3-D locus of each ribbon is found from the intersection of the ray sets from its radiographic images and the analysis of a 3-D spanning tree through the derived ray set intersection points. The seed positions on a ribbon locus are determined by interpolation of equally spaced seed intervals along the 3-D ribbon locus. Experiments with a phantom show that for two radiographs separated by any parallax between 20 degrees and 90 degrees, our technique provides a systematic error in 3-D seed location of 3.0 mm, and a random error of less than 2 mm.

Accumulation of type I fish antifreeze protein in transgenic tobacco is cold-specific.

Expression of fish antifreeze protein (AFP) genes in plants is a possible means of increasing their frost resistance and freeze tolerance. Initial work involved transfer into tobacco of an AFP gene from winter flounder which codes for the alanine-rich, alpha-helical Type I AFP. Plants were transformed with a gene construct in which the preproAFP cDNA was inserted between the cauliflower mosaic virus 19S RNA promoter and the nopaline synthetase polyadenylation site. Although transgenic plants produced AFP mRNA, no AFP was detected on western blots. Re-evaluation of AFP expression in these transgenic plants showed that AFP accumulated to detectable levels only after exposure of the plant to cold. Extracts of plants incubated at 4 degrees C for 24 h contained a protein which co-migrated with winter flounder proAFP and was cross-reactive to Type I AFP antisera. Two other minor protein bands of slightly higher apparent M(r) also cross-reacted with the antisera and are thought to represent processing intermediates. The proAFP was unique to the transgenic plants and was absent in extracts taken prior to cold exposure. AFP levels increased over the first 48 h of cold incubation then remained stable. Since the alpha-helix content of Type I AFP has been shown to decrease markedly at warmer temperatures, we postulate that Type I AFP stability in transgenic plants is dependent on its secondary structure.

The use of clinical parameters in an interactive statistical package to predict pathological features associated with local failure after radical prostatectomy for prostate cancer.

The capability of an interactive statistical package (ISP) to predict the patients whose pathologic findings at the time of radical prostatectomy for prostate cancer would require postoperative radiation therapy to prevent local failure is investigated. A retrospective review of the clinical pretreatment factors and pathologic findings of 174 patients with adenocarcinoma of the prostate treated from 1989 to 1993 with radical retropubic prostatectomy was performed and served as a knowledge base of the ISP. The pathologic findings of seminal vesicle involvement, gross transcapsular disease, and positive surgical margins are defined as outcomes associated with a high risk of local failure after radical prostatectomy and, thereby, requiring postoperative radiation therapy to decrease this risk. By using the pretreatment clinical factors including prostate-specific antigen (PSA), Gleason score, clinical stage, and endorectal magnetic resonance imaging (MRI) findings as input to the ISP, patients are identified from a test group of 50 cases with known pathologic outcome who would require postoperative radiotherapy to decrease local failure. Low- (0% to 33%), intermediate- (34% to 67%), and high-risk groups (68% to 100%) for pathologic features associated with local failure were predicted accurately (r > .95) by the ISP for the 50 test cases. Factors identified on univariate analysis by the ISP as significant predictors of local failure postoperatively include PSA > 20 (p < .001), clinical stage (p < .001), MRI finding of gross transcapsular disease (p < .001), MRI finding of seminal vesicle involvement (p < .001), and Gleason score (p < .003).(ABSTRACT TRUNCATED AT 250 WORDS)

Application of a statistics-based expert system to provide automated second opinions prior to elective surgery.

Second opinions now are mandated by many funding agencies prior to elective surgery to provide an opportunity to evaluate the indications for surgery and allow an evaluation of the risks of surgery. One difficulty with second opinion programs is the observation that 80% to 95% of all surgical recommendations confirm the need for surgery and the number of patients requiring second opinions represents a significant health care cost. The interactive statistical package (ISP) is a statistics-based decision support system currently used to teach residents medical decision processes. Training an ISP is relatively simple because experts need only define the patient parameters and the allowed values of each parameter. Each expert then sets up a decision tree to represent his or her decision process. The system will then present a series of randomly generated cases to the expert and observe the expert's action. Through Bayesian statistics the system identifies the parameter values that significantly affect the expert's judgment and the direction that they exert on the decision. A decision support system used to evaluate patients with prostate cancer reproduces an expert's treatment selection with an accuracy of 94% on a log rank test when selecting among six treatment options including three different surgical procedures, two different radiation regimens, and palliative therapy. The entire system was trained and validated over 3 weeks, and the accuracy of the system is comparable to the consistency of the expert (90%). Because several experts can review the same patient material and make decisions independently, multiple systems can be created and validated, and operated in parallel.(ABSTRACT TRUNCATED AT 250 WORDS)