Predicting Adverse Outcomes in Primary Hyperparathyroidism
During the period 1997–2006, a total of 2299 incident PHPT patients were biochemically identified from electronic databases in Tayside, Scotland, as described in our previous PEARS work, based on a large, relatively unselected population. Of these, <10% (n = 202) were surgically treated. The remaining 2097 untreated PHPT patients were eligible for the study, with a total follow-up of 7338 person years. The mean follow-up was 1278 days (3·5 years) and the maximum was 3644 days (10 years). Overall, 69·9% of the subjects were women, with a mean age 69·5 +/−13·3 years, whilst 30·1% of the subjects were men, with a mean age of 65·7 +/−14·4 years. The baseline biochemical measurements are shown in Table 1. Of all untreated patients, the median serum calcium at baseline was 2·61 mm and the median PTH was 7·2 pm. In total, there were 648 (30·9%) patients who died during the follow-up, 249 of fatal CVD deaths and 182 of cancer-related deaths. Prior to the PHPT diagnosis, 16% had a history of CVD.
When the unadjusted hazards were compared, as shown in Fig. 1a, there was no difference in the risks of developing adverse outcomes between patients with low and high-baseline calcium (P = 0·117, 0·799, 0·882 for mortality, fatal and non-fatal CVD, respectively). The risks, however, were significantly higher in patients with high-baseline PTH compared with those with low PTH for all outcomes observed (Fig.1b, P < 0·001 in all instances). In patients with high-baseline creatinine and ALP, slightly increased risks were also observed (figures not shown).
(Enlarge Image)
Figure 1.
Kaplan-Meier survival plots for mortality, fatal and non-fatal CVD outcomes by baseline serum calcium and plasma PTH separately. Differences between subgroups (high versus low baseline concentrations divided by mean value) are compared by logrank test. (a) Baseline serum calcium (b) Baseline plasma PTH.
For all outcomes examined, the LLS plots were found approximately parallel in categorical variables and thus, the proportional assumption was assumed to be supported for them. For continuous covariates, however, the proportional assumption was clearly violated for baseline calcium and ALP. By examining of the patterns of residual changes over time for those nonproportional variables, a dichotomous time dummy variable indicating whether a case was followed for less than or more than 1000 days was created, as there appeared to be a change at around 1000 days. Thus, a time interaction term, using this dummy time variable together with the nonproportional baseline covariate, was included to address the nonproportionality to improve the specification in the final model. The same approach was taken for baseline PTH and creatinine for consistency reason.
Using the Martingale residuals, the most appropriate functional forms were chosen for all continuous variables. The natural log of baseline PTH, creatinine and ALP and the squared term of calcium (Expressed as Ln(PTH), Ln(CRE), Ln(ALP) and Ca in subsequent tables) was used for further model fitting. The original age variable was retained as linear.
Table 2 shows the HRs of each tested baseline biochemical variables from unadjusted survival models. It can be seen that each of the tested biochemical variables, univariately, was significantly associated with an increased risk in all-cause mortality and fatal CVD, in the short term (< = 1000 days), but such an association diminished over the long term (>1000 days). Baseline serum PTH and creatinine, independently, were risk markers of nonfatal CVD. Serum calcium, univariately, was not a risk factor for nonfatal CVD, nor did serum ALP.
When other covariates were adjusted for, as shown in Table 3, the models fitted much better with smaller AIC and higher C-statistic than the corresponding univariate model. For all-cause mortality, of the biochemical variables tested, only PTH was a consistent risk predictor over the study period (Table 3a). Serum calcium, creatinine and ALP, however, were associated with increased mortality only in the short term, but not over the long term. Previous hospital admission for CVD, cerebrovascular disease, psychiatric condition and cancer also predicted the increased mortality, whereas being a female patient or coming from a lower deprivation decile decreased the risk. For fatal CVD, again, PTH was a biomarker of increased risk over the study period (Table 3b). Serum calcium did not have any significant effect, and the effects of serum creatinine and ALP on the outcome were only significant in the short term. In addition, increasing age, being a male patient, previous usage of bisphosphonates, previous hospital admissions on CVD, cerebrovascular disease and psychiatric condition were also risk factors of fatal CVD. In addition to mortality, of all the tested biochemical indices, baseline PTH was the only significant risk factor for nonfatal CVD. Increasing age, being a male patient and previous hospital admissions on CVD and diabetes also contributed to the increased risk of nonfatal CVD, whilst previous admissions for renal failure and fractures appeared to be risk protectors. For all the outcomes observed, the overall C-statistic of the adjusted models was >0·8, which can be considered as excellent discriminative ability.
Results
Baseline Results
During the period 1997–2006, a total of 2299 incident PHPT patients were biochemically identified from electronic databases in Tayside, Scotland, as described in our previous PEARS work, based on a large, relatively unselected population. Of these, <10% (n = 202) were surgically treated. The remaining 2097 untreated PHPT patients were eligible for the study, with a total follow-up of 7338 person years. The mean follow-up was 1278 days (3·5 years) and the maximum was 3644 days (10 years). Overall, 69·9% of the subjects were women, with a mean age 69·5 +/−13·3 years, whilst 30·1% of the subjects were men, with a mean age of 65·7 +/−14·4 years. The baseline biochemical measurements are shown in Table 1. Of all untreated patients, the median serum calcium at baseline was 2·61 mm and the median PTH was 7·2 pm. In total, there were 648 (30·9%) patients who died during the follow-up, 249 of fatal CVD deaths and 182 of cancer-related deaths. Prior to the PHPT diagnosis, 16% had a history of CVD.
Kaplan–Meier Survival Plots
When the unadjusted hazards were compared, as shown in Fig. 1a, there was no difference in the risks of developing adverse outcomes between patients with low and high-baseline calcium (P = 0·117, 0·799, 0·882 for mortality, fatal and non-fatal CVD, respectively). The risks, however, were significantly higher in patients with high-baseline PTH compared with those with low PTH for all outcomes observed (Fig.1b, P < 0·001 in all instances). In patients with high-baseline creatinine and ALP, slightly increased risks were also observed (figures not shown).
(Enlarge Image)
Figure 1.
Kaplan-Meier survival plots for mortality, fatal and non-fatal CVD outcomes by baseline serum calcium and plasma PTH separately. Differences between subgroups (high versus low baseline concentrations divided by mean value) are compared by logrank test. (a) Baseline serum calcium (b) Baseline plasma PTH.
Proportional Assumption
For all outcomes examined, the LLS plots were found approximately parallel in categorical variables and thus, the proportional assumption was assumed to be supported for them. For continuous covariates, however, the proportional assumption was clearly violated for baseline calcium and ALP. By examining of the patterns of residual changes over time for those nonproportional variables, a dichotomous time dummy variable indicating whether a case was followed for less than or more than 1000 days was created, as there appeared to be a change at around 1000 days. Thus, a time interaction term, using this dummy time variable together with the nonproportional baseline covariate, was included to address the nonproportionality to improve the specification in the final model. The same approach was taken for baseline PTH and creatinine for consistency reason.
Functional Form
Using the Martingale residuals, the most appropriate functional forms were chosen for all continuous variables. The natural log of baseline PTH, creatinine and ALP and the squared term of calcium (Expressed as Ln(PTH), Ln(CRE), Ln(ALP) and Ca in subsequent tables) was used for further model fitting. The original age variable was retained as linear.
Survival Analysis
Table 2 shows the HRs of each tested baseline biochemical variables from unadjusted survival models. It can be seen that each of the tested biochemical variables, univariately, was significantly associated with an increased risk in all-cause mortality and fatal CVD, in the short term (< = 1000 days), but such an association diminished over the long term (>1000 days). Baseline serum PTH and creatinine, independently, were risk markers of nonfatal CVD. Serum calcium, univariately, was not a risk factor for nonfatal CVD, nor did serum ALP.
When other covariates were adjusted for, as shown in Table 3, the models fitted much better with smaller AIC and higher C-statistic than the corresponding univariate model. For all-cause mortality, of the biochemical variables tested, only PTH was a consistent risk predictor over the study period (Table 3a). Serum calcium, creatinine and ALP, however, were associated with increased mortality only in the short term, but not over the long term. Previous hospital admission for CVD, cerebrovascular disease, psychiatric condition and cancer also predicted the increased mortality, whereas being a female patient or coming from a lower deprivation decile decreased the risk. For fatal CVD, again, PTH was a biomarker of increased risk over the study period (Table 3b). Serum calcium did not have any significant effect, and the effects of serum creatinine and ALP on the outcome were only significant in the short term. In addition, increasing age, being a male patient, previous usage of bisphosphonates, previous hospital admissions on CVD, cerebrovascular disease and psychiatric condition were also risk factors of fatal CVD. In addition to mortality, of all the tested biochemical indices, baseline PTH was the only significant risk factor for nonfatal CVD. Increasing age, being a male patient and previous hospital admissions on CVD and diabetes also contributed to the increased risk of nonfatal CVD, whilst previous admissions for renal failure and fractures appeared to be risk protectors. For all the outcomes observed, the overall C-statistic of the adjusted models was >0·8, which can be considered as excellent discriminative ability.
Source...