Association between biochemical control and comorbidities in patients with acromegaly: an Italian longitudinal retrospective chart review study.

PURPOSE: Achieving biochemical control (normalization of insulin-like growth factor-1 [IGF-1] and growth hormone [GH]) is a key goal in acromegaly management. However, IGF-1 and GH fluctuate over time. The true potential impact of time-varying biochemical control status on comorbidities is unclear and relies on multiple, longitudinal IGF-1 and GH measurements. This study assessed the association between time-varying biochemical control status and onset of selected comorbidities in patients with acromegaly. METHODS: Medical charts of adults with confirmed acromegaly and ≥ 6 months of follow-up at an Italian endocrinology center were reviewed. Patients were followed from the first diagnosis of acromegaly at the center until loss to follow-up, chart abstraction, or death. Biochemical control status was assessed annually and defined as IGF-1 ≤ the upper limit of normal, or GH ≤ 2.5 µg/L in the few cases where IGF-1 was unavailable. Time-varying Cox models were used to assess the association between biochemical control status and comorbidities. RESULTS: Among 150 patients, 47% were female, average age at diagnosis was 43.1, and mean length of follow-up was 10.4 years. Biochemical control was significantly associated with a lower hazard of diabetes (HR = 0.36, 95% CI 0.15; 0.83) and cardiovascular system disorders (HR = 0.54, 95% CI 0.31; 0.93), and a higher hazard of certain types of arthropathy (HR = 1.68, 95% CI 1.04; 2.71); associations for other comorbidities did not reach statistical significance. CONCLUSION: Results further support the importance of achieving biochemical control, as this may reduce the risk of high-burden conditions, including diabetes and cardiovascular system disorders. The association for arthropathy suggests irreversibility of this impairment. Due to limitations, caution is required when interpreting these results.

The role of provider characteristics on the hepatitis A and B vaccination status of adults in the United States during 2007-2015.

Hepatitis A and B vaccine coverage is suboptimal in US adults, even among those at increased risk for infection, morbidity, or mortality. To understand where medical education and resources might enhance vaccine coverage, it is important to first identify providers and places most commonly associated with the administration of hepatitis vaccinations. We conducted a retrospective analysis of commercial and Medicare insurance claims data from 2007 to 2015 to describe provider types and places of vaccination against hepatitis A and B among adults in the US, and estimated the time to initial vaccination from first diagnosis of a condition for which the Advisory Committee on Immunization Practices (ACIP) recommends hepatitis A and/or B vaccination among at-risk adults. We identified 183,326 adults who received hepatitis A vaccine, 148,119 hepatitis B vaccine, and 64,953 a bivalent vaccine. Mean age was 42.1-45.8 years. Family practice and internal medicine physicians were the main vaccine providers: 38.9% and 20.2% for hepatitis A, 43.7% and 21.4% for hepatitis B, 35.3% and 15.9% for bivalent vaccinations, respectively. ≥90% of initial vaccinations occurred in an office practice. In at-risk patients, median time to first-dose received was 11.8, 20.9, and 20.9 months for hepatitis A, hepatitis B, and hepatitis A/B vaccines, respectively. Primary care and office practices were the most common providers and places of vaccination, respectively, for hepatitis A and B vaccine. For at-risk patients, further research is needed to design vaccination strategies to improve the median time from first ACIP-recommended condition diagnosis to initial vaccination against hepatitis A and B.

Impact of symptomatic skeletal events on health-care resource utilization and quality of life among patients with castration-resistant prostate cancer and bone metastases.

BACKGROUND: Data regarding the impact of symptomatic skeletal events (SSEs) on health economics and patient-reported outcomes in men with castration-resistant prostate cancer (CRPC) and bone metastases from a clinical setting are lacking. Hence, this study aimed to quantify the effects of SSEs on health-care resource utilization (HRU), health-related quality of life (HRQoL) and pain in men with CRPC metastasized to bone. METHODS: This cohort study included men with CRPC and bone metastasis treated at a tertiary center during December 1996-July 2015. SSEs, including pathological fracture, radiation to bone, spinal cord compression and bone surgery, as well as HRU were identified retrospectively through medical records and clinical database. A subset of surviving patients completed Functional Assessment of Cancer Therapy-Prostate (FACT-P) and Brief Pain Inventory-Short Form (BPI-SF) questionnaires. The incremental effect of SSEs on HRU was evaluated using multivariable generalized linear regression. Questionnaire scores were compared using effect sizes (ES); ES⩾0.33 indicated meaningful differences between SSE and non-SSE cohorts. Lower scores suggest lower HRQoL and pain. RESULTS: Of the 832 patients, 207 developed ⩾1 SSE (mean 1.5±0.8) during follow-up (median 2.1 years). Radiation to bone was the most common SSE (84.1%). SSE cohort had significantly higher emergency room (incidence rate ratio (IRR)=1.48; P=0.006), outpatient (IRR=1.17; P=0.005) and inpatient (IRR=1.74; P<0.001) visits. Of the 107 eligible survey patients, 103 (96.3%) responded. SSE cohort had lower mean FACT-P functional well-being (17.5 vs 19.8; P=0.158; ES=0.36), higher mean pain severity (2.5 vs 1.6; P=0.048; ES=0.47) and worst pain scores (3.6 vs 2.3; P=0.033; ES=0.50) compared with the non-SSE cohort, indicating meaningful differences between cohorts. CONCLUSIONS: This study demonstrated high economic and HRQoL burden of SSEs. The findings underscore the need for better supportive and disease-modifying treatments for these patients.

In-hospital risk of venous thromboembolism and bleeding and associated costs for patients undergoing total hip or knee arthroplasty.

OBJECTIVE: Benefits of anti-coagulation for venous thromboembolism (VTE) prevention in total hip and knee arthroplasty (THA/TKA) may be offset by increased risk of bleeding. The aim was to assess in-hospital risk of VTE and bleeding after THA/TKA and quantify any increased costs. METHODS: Healthcare claims from the Premier Perspective(TM) Comparative Hospital Database (January 2000-September 2008) were selected for subjects ≥ 18 years with ≥ 1 diagnosis code for THA/TKA. VTE was defined as ≥ 1 code for deep vein thrombosis or pulmonary embolism. Bleeding was classified as major/non-major. Incremental in-hospital costs associated with VTE and bleeding were calculated as cost differences between inpatients with VTE or bleeding matched 1:1 with inpatients without VTE or bleeding. RESULTS: A total of 820,197 inpatient stays were identified: 8042 had a VTE event and 7401 a bleeding event (2740 major bleeding). The risks of VTE, any bleeding, and major bleeding were 0.98, 0.90, and 0.33/100 inpatient stays, respectively. Mean incremental in-hospital costs per inpatient were $2663 for VTE, $2028 for bleeding, and $3198 for major bleeding. LIMITATIONS: These included possible inaccuracies or omissions in procedures, diagnoses, or costs of claims data; no information on the amount of blood transfused or decreases in the hemoglobin level to evaluate bleeding event severity; and potential biases due to the observational design of the study. CONCLUSIONS: In-hospital risk and incremental all-cause costs with THA/TKA were higher for VTE than for bleeding. Despite higher costs, major bleeding occurred less frequently than VTE, suggesting a favorable benefit/risk profile for VTE prophylaxis in THA/TKA.

Persistence and compliance of deferoxamine versus deferasirox in Medicaid patients with sickle-cell disease.

WHAT IS KNOWN AND OBJECTIVE: Patients with sickle-cell disease (SCD) receiving chronic transfusions of red blood cells are at risk of developing serious adverse effects. Iron chelation therapy (ICT) helps eliminate iron overload by binding with plasma iron to form a non-toxic conjugate that can be safely excreted from the body. Two iron chelating agents are currently available in the United States: Deferoxamine (DFO) is an injectable formulation, and deferasirox (Exjade(®) ) is an oral suspension. This study compared the frequency of hospitalizations, persistence and compliance of patients with SCD from Medicaid programmes treated with DFO vs. deferasirox. METHODS: Health care claims from Medicaid Florida (1998-2007), Missouri (1993-2008) and New Jersey (1996-2008) were analysed. Patients with continuous enrolment for ≥6months prior to ICT initiation and ≥1 SCD diagnosis were included in the analysis. Patients were divided into four cohorts: patients treated with DFO (any-DFO group) and patients treated with deferasirox (any-deferasirox group); the latter was further divided into patients initiated on DFO and then switched to deferasirox (deferasirox switchers), and patients treated with deferasirox-only (deferasirox-only group). Frequency of hospitalization for crisis conditions related to SCD as well as length of stay pre- and post-ICT treatment initiation were assessed. Persistence was defined as time to drug discontinuation with ≥1 Rx gap, using Kaplan-Meier approach. Compliance was estimated using a medication possession ratio (MPR) based on the drug exposure approach. Adjusted analyses of persistence and compliance were also conducted. RESULTS: A total of 217 (mean age: 19·4years, 39·2 men), 275 (20·1years, 41·5% men), 105 (19·4years, 42·9% men) and 166 (20·4years, 41·6% men) patients were included in the any-DFO, any-deferasirox, deferasirox switchers and deferasirox-only groups, respectively. After ICT initiation, the any-deferasirox and deferasirox-only groups experienced a statistically significant reduction in the frequency of hospitalizations relative to pretreatment [any-deferasirox: from 0·09 to 0·06 hospitalizations per patient per month (pmpm), P=0·0105; deferasirox-only: from 0·11 to 0·07 hospitalizations pmpm, P=0·0188], whereas it remained stable in the any-DFO group at 0·08 hospitalizations pmpm (P=0·9483). The Kaplan-Meier rates of medication persistence assessed at 6 and 12months of follow-up were significantly lower for DFO patients (6 months: 0·34, 12months: 0·21) as compared to all deferasirox (0·51, 0·29, P=0·0002), deferasirox switchers (0·56, 0·37, P=0·0002) and deferasirox-only (0·47, 0·24, P=0·0176) patients. Similarly, compliance to treatment was significantly lower for patients treated with DFO (mean MPR: 0·64) compared with any-deferasirox (0·78, P<0·0001), deferasirox switchers (0·75, P=0·0002) and deferasirox-only (0·80, P<0·0001) patients. Adjusted analyses of persistence and compliance yielded similar results. WHAT IS NEW AND CONCLUSIONS: Based on a Medicaid population, patients treated with deferasirox were more compliant and persistent with their treatment than those treated with DFO. Frequency of hospitalizations was significantly reduced after treatment initiation for the any-deferasirox and deferasirox-only groups. Prospective studies controlling for potential clinical and treatment pattern differences between deferasirox and DFO patients are needed to assess whether the decreased hospitalizations after initiation of deferasirox are related to better treatment compliance.

Health-related quality of life in patients with advanced renal cell carcinoma receiving pazopanib or placebo in a randomised phase III trial.

BACKGROUND: In a double-blind, randomised phase III trial of advanced renal cell carcinoma patients, pazopanib 800mg QD (n=290) versus placebo (n=145) significantly prolonged progression-free survival (hazard ratio (HR)=0.46, 95% confidence interval [CI] 0.34-0.62, p-value<0.0001), without important differences in health-related quality of life (HRQoL). This post-hoc analysis evaluated time to HRQoL deterioration and whether tumour response/stabilisation was associated with HRQoL improvement. METHODS: HRQoL was assessed using EORTC QLQ-C30 and EQ-5D. Effect of pazopanib on time to ⩾20% decline from baseline in summary scores was estimated for all patients and by prior treatment. Analyses were conducted for different HRQoL deterioration thresholds. HRQoL changes were stratified by benefit and compared: complete response (CR) or partial response (PR) versus progressive disease (PD); CR/PR versus stable disease (SD), and SD versus PD. RESULTS: There was a trend for pazopanib patients to be less likely than placebo patients to experience ⩾20% HRQoL deterioration in EORTC-QLQ-C-30 global health status/QOL scale (HR=0.77; 95% CI 0.57-1.03, not significant). Results by prior treatment and different HRQoL deterioration thresholds were similar. Patients with CR/PR and SD experienced significantly less HRQoL deterioration than those with PD (p<0.001, p=0.0024, respectively); mean differences between patients with CR/PR and PD exceeded the pre-determined minimally important difference (MID). Differences between patients with SD and PD did not exceed pre-determined MID. Results were generally consistent across treatment and EQ-5D summary scores. CONCLUSION: Results support the favourable benefit-risk profile of pazopanib and suggest patients experiencing tumour response/stabilisation also may have better HRQoL compared to those without this response.

Generic antiepileptic drugs and associated medical resource utilization in the United States.

OBJECTIVE: To evaluate whether generic substitution was associated with any difference in medical resource utilization for 5 widely used antiepileptic drugs (AEDs) in the United States. METHODS: Health insurance claims from PharMetrics Database, representing over 90 health plans between January 2000 and October 2007, were analyzed. Adult patients with epilepsy, continuously treated with carbamazepine, gabapentin, phenytoin, primidone, or zonisamide, were selected. An open-cohort design was used to classify patients into mutually exclusive periods of brand vs generic use of AEDs. Pharmacy and medical utilization were compared between the 2 periods with multivariate regression analyses. Results were stratified into epilepsy-related medical services, and stable (< or = 2 outpatient visits per year and no emergency room visit) vs unstable epilepsy. Time-to-event analyses were also performed for all services and epilepsy-related endpoints. RESULTS: A total of 18,125 patients were observed in the stable group and 15,500 patients in the unstable group. After adjustment of covariates, periods of generic AED treatment were associated with increased use of all prescription drugs (incidence rate ratio [IRR] [95% confidence interval (CI)] = 1.13 [1.13-1.14]) and higher epilepsy-related medical utilization rates (hospitalizations: IRR [95% CI] = 1.24 [1.19-1.30]; outpatient visits: IRR [95% CI] = 1.14 [1.13-1.16]; lengths of hospital stays: IRR [95% CI] = 1.29 [1.27-1.32]). Generic-use periods were associated with increased utilization rates in stable and unstable patients and with 20% increased risk of injury, compared to periods with brand use of AEDs. CONCLUSIONS: Generic antiepileptic drug use was associated with significantly greater medical utilization and risk of epilepsy-related medical events, compared to brand use. This relationship was observed even in patients characterized as stable. AED = antiepileptic drug; CI = confidence interval; ER = emergency room; HR = hazard ratio; ICD = International Classification of Diseases; IRR = incidence rate ratio.

The risks and costs of multiple-generic substitution of topiramate.

OBJECTIVE: To investigate clinical and economic consequences following generic substitution of one vs multiple generics of topiramate (Topamax; Ortho-McNeil Neurologics, Titusville, NJ). METHODS: Medical and pharmacy claims data of Régie de l'Assurance-Maladie du Québec from January 2006 to October 2007 were used. Patients with epilepsy treated with topiramate were selected. An open-cohort design was used to classify the observation period into periods of brand, single-generic, and multiple-generic use. One-year generic-switch and switchback-to-brand rates were estimated using Kaplan-Meier methodology. Medical resource utilization and costs were compared among the three periods using multivariate regression analysis. RESULTS: In total, 948 patients were observed during 1,105 person-years of brand use, 233 person-years of single-generic use, and 92 person-years of multiple-generic use. A total of 23% of generic users received at least two different generic versions. Compared to brand use, multiple-generic use was associated with higher utilization of other prescription drugs (incidence rate ratio [IRR] = 1.27, 95% confidence interval [CI] = 1.24-1.31), higher hospitalization rates (0.48 vs 0.83 visit/person-year, IRR = 1.65, 95% CI = 1.28-2.13), and longer hospital stays (2.6 vs 3.9 days/person-year, IRR = 1.43, 95% CI = 1.27-1.60), but the effect was less pronounced in single-generic use (hospitalization: IRR = 1.08, 95% CI = 0.88-1.34, length of stay: IRR = 1.12, 95% CI = 1.03-1.23). The risk of head injury or fracture was nearly three times higher (hazard ratio = 2.84, 95% CI = 1.24-6.48) following a generic-to-generic switch compared to brand use. The total annualized health care cost per patient was higher in the multiple-generic than brand periods by C$1,716 (cost ratio = 1.21, p = 0.0420). CONCLUSION: Multiple-generic substitution of topiramate was significantly associated with negative outcomes, such as hospitalizations and injuries, and increased health care costs.

Nonadherence to antiepileptic drugs and increased mortality: findings from the RANSOM Study.

OBJECTIVES: The primary objective was to investigate whether nonadherence to antiepileptic drugs (AEDs) is associated with increased mortality and the secondary objective to examine whether nonadherence increases the risk of serious clinical events, including emergency department (ED) visits, hospitalizations, motor vehicle accident (MVA) injuries, fractures, and head injuries. METHODS: A retrospective open-cohort design was employed using Medicaid claims data from Florida, Iowa, and New Jersey from January 1997 through June 2006. Patients aged > or =18 years with > or =1 diagnosis of epilepsy by a neurologist and > or =2 AED pharmacy dispensings were selected. Medication possession ratio (MPR) was used to evaluate AED adherence on a quarterly basis with MPR > or =0.80 considered adherent and <0.80 nonadherent. The association of nonadherence with mortality was assessed using a time-varying Cox regression model adjusting for demographic and clinical confounders. Incidence rates for serious clinical events were compared between adherent and nonadherent quarters using incidence rate ratios (IRRs) with 95% CIs calculated based on the Poisson distribution. RESULTS: The 33,658 study patients contributed 388,564 AED-treated quarters (26% nonadherent). Nonadherence was associated with an over threefold increased risk of mortality compared to adherence (hazard ratio = 3.32, 95% CI = 3.11-3.54) after multivariate adjustments. Time periods of nonadherence were also associated with a significantly higher incidence of ED visits (IRR = 1.50, 95% CI = 1.49-1.52), hospital admissions (IRR = 1.86, 95% CI = 1.84-1.88), MVA injuries (IRR = 2.08, 95% CI = 1.81-2.39), and fractures (IRR = 1.21, 95% CI = 1.18-1.23) than periods of adherence. CONCLUSION: These findings suggest that nonadherence to antiepileptic drugs can have serious or fatal consequences for patients with epilepsy.

Clinical consequences of generic substitution of lamotrigine for patients with epilepsy.

OBJECTIVES: To measure the proportions of patients switching from generic to branded drugs among users of antiepileptic drugs (AED) compared to other therapeutic areas and to investigate medical services utilization associated with generic switching of lamotrigine. METHODS: Medical and pharmacy claims data from Régie de l'Assurance Maladie du Québec database from April 1998 to July 2006 were used. Patients with an epilepsy diagnosis (International Classification of Diseases-9 345) and treated with lamotrigine for >60 of the 90 days before the entry date of generic lamotrigine in Quebec (February 1, 2003) were selected. The proportion of patients switching back to brand were calculated for lamotrigine, for other AEDs (clobazam, carbamazepine CR, gabapentin) and for non-AED chronic medications (carvedilol, fosinopril, simvastatin). Medical resource utilization was compared between periods of branded vs generic use of lamotrigine. RESULTS: Of 671 patients treated with branded lamotrigine, 187 patients (27.9%) switched to a generic, and 51 of these patients (27.5%) switched back to the branded medication. Rates of switchback were from 20.8% to 44.1% for various AEDs and from 7.7% to 9.1% for non-AEDs. Relative to the branded lamotrigine use period, generic lamotrigine use period was associated with a 5.1% increase in mean daily dose of lamotrigine (239.1 vs 251.4 mg; p = 0.0149), a higher number of dispensations for other AEDs (20.4 vs 23.9 dispensations per person-year; p < 0.001) as well as non-AED drugs (26.4 vs 32.8 dispensations per person-year; p < 0.0001), a higher utilization rate of medical services (8.7 vs 9.8 visits per person-year; p < 0.0001), and a longer hospital length of stay (3.29 days vs 4.86 days per person-year; p < 0.0001). CONCLUSION: A higher propensity to switch back to branded medications was observed among antiepileptic drug users compared to users of antihypertensives and antihyperlipidemics, similar to findings from Andermann et al. Switch to generic lamotrigine was significantly associated with increased physician visits and hospitalizations.

The cost-effectiveness of weekly epoetin alfa relative to weekly darbepoetin alfa in patients with chemotherapy-induced anemia.

OBJECTIVE: To compare the cost-effectiveness of epoetin alfa (EPO) and darbepoetin alfa (DARB) for the treatment of chemotherapy-induced anemia (CIA), using dosing regimens approved by the FDA (EPO 40,000 U once weekly and DARB 2.25 U once weekly and DARB 2.25 mcg/kg once weekly). METHODS: The study compared published results of two double-blind, randomized, phase III trials one utilizing EPO (N = 166) and the other, DARB (N = 367). Patients in both trials similar baseline characteristics. Effectiveness was measured as the proportion of EPO or DARB patients who were successfully treated (i.e., did not require blood transfusion) during weeks 0-16 and 5-16, respectively. Estimated drug costs were presented in 2005 USD based on wholesale acquisition cost (WAC) and average drug utilization over 16 weeks. Cost-effectiveness was calculated as the estimated drug costs divided by transfusion effectiveness. Threshold analysis was used to determine the break-even point at which EPO and DARB had the same drug costs. RESULTS: Estimated drug costs over 16 weeks were $9,039 for EPO and $13,555 for DARB. During weeks 5-16, 85% of EPO patients and 73% of DARB patients were successfully treated, resulting in average cost-effectiveness ratios of $106 for EPO and $186 for DARB per one per cent of successfully treated patients. A 33% reduction in DARB WAC was required to achieve the same drug costs as for EPO. CONCLUSIONS: Utilizing FDA-approved doses, EPO was found to result in lower drug costs and better treatment success when compared to DARB. Hence, EPO is a dominant alternative compared to DARB for the treatment of CIA. The analyses presented here are not without limitations. Specifically, although the studies were comparable, patients were ultimately drawn from different populations.

Association between intraocular pressure and budesonide inhalation therapy in asthmatic patients.

BACKGROUND: The extent to which inhaled glucocorticoids increase the risk of intraocular pressure elevation has been controversial. OBJECTIVE: The authors attempt to assess such risk attributable to budesonide, an inhaled glucocorticoid for asthma therapy. METHODS: Data were pooled from four prospective, randomized, double-blind, parallel-group, placebo-controlled, multicenter clinical trials of 12 to 20 weeks in duration. One thousand two hundred and fifty-five patients, 6 to 70 years of age whose intraocular pressures (IOPs) were less than 23 mmHg at screening were randomized to receive placebo or inhaled budesonide at doses ranging from 100 to 800 microg, administered twice daily. Intraocular pressure was measured at screening and at the end of double-blind treatment. Intraocular change was compared between budesonide and placebo, accounting for the confounding effects of gender, race, age, history of diabetes, history of hypertension, clinical trial, systemic glucocorticoid use during the trials, ophthalmic glucocorticoid use during the trials, and prior oral glucocorticoid use. RESULTS: No budesonide treatment effect on the IOP was evident either in the crude analysis or after adjustment for possible confounding factors. For patients exposed to budesonide at a total daily dose of 1600 microg for 20 weeks, there was no difference in IOP change compared with the placebo controls. CONCLUSIONS: No association with an increased IOP was observed in asthmatic patients treated with budesonide at daily doses ranging from 200 to 1600 microg for durations of 12 to 20 weeks. The subgroup analysis, which focused on the highest dose and longer term therapy was reassuring, as was the overall result.

Descriptive epidemiology of acute liver enzyme abnormalities in the general population of central Massachusetts.

OBJECTIVES: This study quantifies the incidence of liver enzyme abnormalities in the general population of central Massachusetts. METHODS: Computerized data files from a health maintenance organization were used to ascertain potential subjects during the study period of 1 July 1992-30 June 1993. Medical records and laboratory tests results were reviewed to confirm the diagnoses. RESULTS: The incidence rates were 40.6 cases per 100,000 persons per year for drug-associated liver enzyme abnormalities, 39.0 per 100,000 persons per year for liver enzyme abnormalities secondary to biliary pathologies, 25.2 per 100,000 persons per year for liver enzyme abnormalities secondary to mononucleosis, 25.2 per 100,000 persons per year for liver enzyme abnormalities of unknown aetiology, 15.4 per 100,000 persons per year for alcoholic liver enzyme abnormalities, 12.2 per 100,000 persons per year for liver malignancy, and 7.3 per 100,000 persons per year for viral hepatitis (HAV, HBV and HCV). Men were more commonly affected by alcoholic liver enzyme abnormalities and viral hepatitis, with respective incidence rates approximately 5 and 2 times higher than women. Liver enzyme abnormalities secondary to mononucleosis occurred predominantly between the ages of 15-24 years. In contrast, liver diseases secondary to biliary pathologies, liver malignancies and drug-associated liver enzyme abnormalities were found most frequently among the elderly. CONCLUSION: Liver enzyme abnormalities occurred with a higher incidence in this general population than reported previously in selected patients. Drug-associated liver enzyme abnormality was the most common type. The magnitude of drug-associated liver enzyme abnormality could be much higher as an appreciable proportion of the liver cases of unknown aetiology are potentially drug-attributable.

Epidemiologic interpretation of artificial neural networks.

Multilayer neural networks have been faulted for functioning as "black boxes" and for failing to assess the relative importance of the input factors. The aim of this paper is to illustrate how neural networks can classify individuals. The authors investigated the role of weights in the formation of neural networks' decision surfaces and decision regions. The data used were from a case-control study. Two strong determinants of case status were used as input "neurons." Zero, three, and five hidden neurons were used to explore the effect of the number of hidden neurons on the decision surfaces and regions. Mapping of input and output spaces revealed that three hidden neurons were insufficient to fully discriminate cases from controls. Five hidden neurons may be optimal, but at the cost of possible over-fitting. The more complex neural networks were very effective at defining regions of uniform risk in the plane of the initial covariates, and at assigning risk levels. The authors speculate that neural networks will prove useful in epidemiologic problems that require pattern recognition or complicated classification techniques, and that they will be unfavorable in problems that involve distinct effects of distinguishable predictors.

Prediction and cross-validation of neural networks versus logistic regression: using hepatic disorders as an example.

The authors developed and cross-validated prediction models for newly diagnosed cases of liver disorders by using logistic regression and neural networks. Computerized files of health care encounters from the Fallon Community Health Plan were used to identify 1,674 subjects who had had liver-related health services between July 1, 1992, and June 30, 1993. A total of 219 subjects were confirmed by review of medical records as incident cases. The 1,674 subjects were randomly and evenly divided into training and test sets. The training set was used to derive prediction algorithms based solely on the automated data; the test set was used for cross-validation. The area under the Receiver Operating Characteristic curve for a neural network model was significantly larger than that for logistic regression in the training set (p = 0.04). However, the performance was statistically equivalent in the test set (p = 0.45). Despite its superior performance in the training set, the generalizability of the neural network model is limited. Logistic regression may therefore be preferred over neural network on the basis of its established advantages. More generalizable modeling techniques for neural networks may be necessary before they are practical for medical research.

The effectiveness of surgery on the treatment of acute spinal cord injury and its relation to pharmacological treatment.

Using data from the Second National Acute Spinal Cord Injury Study (NASCIS II), the authors sought to characterize the role of surgery in the management of traumatic spinal cord injury and to examine the interaction between pharmacological treatment and surgery. Patients who did not undergo surgery had more severe spinal cord injuries initially than those who had surgery. However, no differences in neurological improvement at 1-year follow-up were found between those who underwent surgery and those who did not. The results suggest that either early surgery (< or = 25 hours after injury) or late surgery (> 200 hours) may be associated with increased neurological recovery, particularly motor function, but these results are equivocal. Surgery was not shown to interact with pharmacological treatments, indicating that the effect of drug treatment in NASCIS II, reported elsewhere, is not influenced by surgery. Other independent variables that best predicted improvement in motor score were age of 25 years or younger, incomplete injury, and lower baseline emergency department neurological scores. This study does not provide clinically relevant evidence concerning the efficacy of timing or the value of surgery in treating patients with spinal cord injuries. A randomized study on the timing and efficacy of spinal cord surgery is needed to obtain valid comparisons of the efficacy of surgical treatments.