Why does this piecewise linear mixed model not produce equal estimates at the knot - ggplot2

Question

I am wondering if someone could help me interpret my piecewise lmm results. Why does ggpredict() produce different estimates for the knot at 10 weeks (end of tx; see ‘0’ in graph at end)? I've structured the data like so:
bpiDat <- bpiDat %>%
mutate(baseToEndTx = ifelse(week <= 10, week, 1)) %>%
mutate(endOfTxToFu = case_when(
week <= 10 ~ 0,
week == 18 ~ 8,
week == 26 ~ 16,
week == 34 ~ 24
)) %>%
select(id, treatment, baseHamd, week, baseToEndTx, endOfTxToFu,
painInterferenceMean, painSeverityMean, bpiTotal) %>%
mutate(baseHamd = scale(baseHamd, scale=F))
Which looks like this:
id treatment baseHamd week baseToEndTx endOfTxToFu painSeverityMean
1 1 4.92529343 0 0 0 6.75
1 1 4.92529343 2 2 0 7.25
1 1 4.92529343 4 4 0 8.00
1 1 4.92529343 6 6 0 NA
1 1 4.92529343 8 8 0 8.25
1 1 4.92529343 10 10 0 8.00
1 1 4.92529343 18 1 8 8.25
1 1 4.92529343 26 1 16 8.25
1 4.92529343 34 1 24 8.00
The best fitting model:
model8 <- lme(painSeverityMean ~ baseHamd + baseToEndTx*treatment + endOfTxToFu + I(endOfTxToFu^2)*treatment,
data = bpiDat,
method = "REML",
na.action = "na.exclude",
random = ~baseToEndTx | id)
This is how I’m visualizing:
test1 <- ggpredict(model8, c("baseToEndTx", "treatment"), ci.lvl = NA) %>%
mutate(x = x - 10) %>%
mutate(phase = "duringTx")
test2 <- ggpredict(model8, c("endOfTxToFu", "treatment"), ci.lvl = NA) %>%
mutate(phase = "followUp")
t <- rbind(test1, test2)
t <- t %>%
pivot_wider(names_from = "phase",
values_from = "predicted")
ggplot(t) +
geom_smooth(aes(x,duringTx,col=group),method="lm",se=FALSE) +
geom_smooth(aes(x,followUp,col=group),method="lm",se=FALSE) +
geom_point(aes(x,duringTx,col=group)) +
geom_point(aes(x,followUp,col=group)) +
ylim(2,6)
Which produces this: